Datasets:
tag stringlengths 23 41 | category stringclasses 6
values | subcategory stringlengths 5 21 | difficulty stringclasses 4
values | prompt_1 stringlengths 37 332 | prompt_2 stringlengths 36 251 | prompt_3 stringlengths 37 318 | question_1 stringlengths 91 479 | question_2 stringlengths 88 456 | question_3 stringlengths 87 496 | evaluation_criteria stringlengths 17 59 | vision_question_template stringlengths 149 306 | vlm stringclasses 5
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
TextRendering_WritingAccuracy_Easy | Text Rendering | Writing accuracy | Easy | A single large word 'HELLO' centered on a white background | The word 'ADVENTURE' in big bold letters centered on a plain cream background | The word 'SERENITY' written large and centered on a solid light blue background | Does the image show the single word "HELLO", correctly spelled and legible? PASS if the text reads exactly HELLO; FAIL if misspelled, garbled, or different. | Does the image show the single word "ADVENTURE", correctly spelled and legible? PASS if the text reads exactly ADVENTURE; FAIL if misspelled, garbled, or different. | Does the image show the single word "SERENITY", correctly spelled and legible? PASS if the text reads exactly SERENITY; FAIL if misspelled, garbled, or different. | Word spelled correctly — binary yes/no | Look at this image. The intended text was: "{prompt_excerpt}". Is every letter spelled correctly and fully readable? List any misspellings, missing letter, or illegible parts. Rate accuracy: PASS if all text is correct, FAIL otherwise. | qwen3-vl |
TextRendering_WritingAccuracy_Medium | Text Rendering | Writing accuracy | Medium | The sentence 'The quick brown fox jumps high' in clean black text on a white background | 'Every sunset brings a new dawn ahead' written neatly on a pale yellow background | 'Stars shine brightest in the dark night' displayed on a clean off-white background | Does the image show the exact text "The quick brown fox jumps high", every word spelled correctly and fully readable? PASS if the text reads exactly that with no misspellings, missing, or extra words; FAIL otherwise. | Does the image show the exact text "Every sunset brings a new dawn ahead", every word spelled correctly and fully readable? PASS if it reads exactly that with no misspellings, missing, or extra words; FAIL otherwise. | Does the image show the exact text "Stars shine brightest in the dark night", every word spelled correctly and fully readable? PASS if it reads exactly that with no misspellings, missing, or extra words; FAIL otherwise. | All words correct, readable | Look at this image. The intended text was: "{prompt_excerpt}". Is every word spelled correctly and fully readable? List any misspellings, missing words, or illegible parts. Rate accuracy: PASS if all text is correct, FAIL otherwise. | qwen3-vl |
TextRendering_WritingAccuracy_Hard | Text Rendering | Writing accuracy | Hard | A poster with headline 'SUMMER SALE' in large bold text, body text 'Up to 50% off all items this weekend only' in medium size, and fine print 'Terms and conditions apply. See store for details.' at the bottom in small text | A flyer with large title 'GRAND OPENING', subtitle 'Join us for food, music, and fun' in medium text, and small disclaimer 'RSVP required. Limited seating available.' at the bottom | An event banner with headline 'ANNUAL GALA 2025' at the top in large serif font, description 'An evening of fine dining and live jazz' in medium text, and footnote 'Black tie optional. Doors open at 6 PM.' in small text | Does the image show a poster containing the headline "SUMMER SALE", the body line "Up to 50% off all items this weekend only", and the fine print "Terms and conditions apply. See store for details."? PASS only if all three texts are present, correctly spelled, and fully readable; FAIL if any is missing, misspelled, or ... | Does the image show a flyer containing the title "GRAND OPENING", the subtitle "Join us for food, music, and fun", and the disclaimer "RSVP required. Limited seating available."? PASS only if all three texts are present, correctly spelled, and fully readable; FAIL if any is missing, misspelled, or illegible. | Does the image show an event banner containing the headline "ANNUAL GALA 2025", the description "An evening of fine dining and live jazz", and the footnote "Black tie optional. Doors open at 6 PM."? PASS only if all three texts are present, correctly spelled, and fully readable; FAIL if any is missing, misspelled, or i... | All text correct + visual hierarchy respected | Look at this image. The intended text was: "{prompt_excerpt}". Is every word spelled correctly and fully readable? List any misspellings, missing words, or illegible parts. Rate accuracy: PASS if all text is correct, FAIL otherwise. | qwen3-vl |
TextRendering_WritingAccuracy_Extreme | Text Rendering | Writing accuracy | Extreme | A vintage movie poster for 'THE LAST VOYAGE' featuring a tagline 'One ship. One chance. No turning back.' with credits block reading 'Directed by James Harlow. Starring Elena Voss, Marcus Chen, and Sofia Reyes. Music by Thomas Grant. A Paramount Pictures Release. Coming Summer 2025.' integrated realistically into the p... | A magazine cover titled 'NATIONAL EXPLORER' with cover lines 'The Hidden Temples of Cambodia — page 34', 'Interview: Dr. Maria Santos on Climate Action', 'Top 10 Trails You Must Hike This Year', and a barcode area with 'Issue 247 | March 2025 | $6.99' | A product label for 'GOLDEN HARVEST ORGANIC HONEY' with subtitle 'Raw & Unfiltered', weight '12 oz (340g)', description 'Pure wildflower honey harvested from sustainable apiaries in the Pacific Northwest. No additives or preservatives.', and small text 'Best before: 06/2026. Store in cool, dry place. Product of USA.' | Does the image show a movie poster with the title "THE LAST VOYAGE", the tagline "One ship. One chance. No turning back.", and the credits block "Directed by James Harlow. Starring Elena Voss, Marcus Chen, and Sofia Reyes. Music by Thomas Grant. A Paramount Pictures Release. Coming Summer 2025."? PASS only if all of th... | Does the image show a magazine cover with the title "NATIONAL EXPLORER" and the cover lines "The Hidden Temples of Cambodia — page 34", "Interview: Dr. Maria Santos on Climate Action", "Top 10 Trails You Must Hike This Year", and "Issue 247 | March 2025 | $6.99"? PASS only if all of this text is present, correctly spel... | Does the image show a honey product label with the title "GOLDEN HARVEST ORGANIC HONEY", subtitle "Raw & Unfiltered", weight "12 oz (340g)", the description "Pure wildflower honey harvested from sustainable apiaries in the Pacific Northwest. No additives or preservatives.", and the small text "Best before: 06/2026. Sto... | Complete text accuracy + realistic integration | Look at this image. The intended text was: "{prompt_excerpt}". Is every word spelled correctly and fully readable? List any misspellings, missing words, or illegible parts. Rate accuracy: PASS if all text is correct, FAIL otherwise. | qwen3-vl |
TextRendering_TypographyStyle_Easy | Text Rendering | Typography Style | Easy | The word 'CHAPTER ONE' typed on aged paper with a vintage typewriter font, complete with slightly uneven ink | The phrase 'TODAY'S SPECIAL' written in white chalk on a dark green chalkboard with chalk dust | The word 'OPEN' displayed on a red LED sign with glowing pixel-style letters | Does the image show the text "CHAPTER ONE" rendered in a vintage typewriter-style font on aged paper? PASS if the text reads exactly CHAPTER ONE, correctly spelled and legible, in a typewriter look on aged paper; FAIL if the text is wrong or the style/medium does not match. | Does the image show the phrase "TODAY'S SPECIAL" written in white chalk on a dark green chalkboard? PASS if the text reads exactly TODAY'S SPECIAL, correctly spelled and legible, in a chalk style on a chalkboard; FAIL if the text is wrong or the style/medium does not match. | Does the image show the word "OPEN" displayed as glowing pixel-style letters on a red LED sign? PASS if the text reads exactly OPEN, correctly spelled and legible, in a glowing red LED style; FAIL if the text is wrong or the style/medium does not match. | Style + material match | This image should show text rendered in a specific style/medium. The prompt was: "{prompt_excerpt}". Is every word spelled correctly and fully readable? List any misspellings, missing words, or illegible parts. Rate accuracy. Additionally does the text style match the described medium? Rate: PASS or FAIL. | qwen3-vl |
SpatialReasoning_Counting_Easy | Spatial Reasoning | Counting | Easy | Exactly 3 red apples arranged on a wooden table | Exactly 5 yellow rubber ducks sitting in a row on a bathroom shelf | Exactly 4 white candles standing on a dark surface | How many red apples are on the table? PASS if there are exactly 3 red apples; FAIL for any other number. | How many yellow rubber ducks are on the shelf? PASS if there are exactly 5; FAIL for any other number. | How many white candles are standing on the surface? PASS if there are exactly 4; FAIL for any other number. | Count correct — binary | Count the objects in this image. The prompt asked for: "{prompt_excerpt}". How many of each object type do you see? Does the count exactly match what was requested? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_Counting_Medium | Spatial Reasoning | Counting | Medium | 2 orange cats and 3 black dogs sitting together on a grassy lawn | 4 red roses and 2 white daisies in a glass vase on a table | 1 blue butterfly and 3 yellow butterflies resting on a garden fence | Count the animals: how many orange cats and how many black dogs are there? PASS only if there are exactly 2 orange cats and 3 black dogs; FAIL if either count differs. | Count the flowers: how many red roses and how many white daisies are in the vase? PASS only if there are exactly 4 red roses and 2 white daisies; FAIL if either count differs. | Count the butterflies: how many blue butterflies and how many yellow butterflies are there? PASS only if there is exactly 1 blue butterfly and 3 yellow butterflies; FAIL if either count differs. | Both counts correct | Count the objects in this image. The prompt asked for: "{prompt_excerpt}". How many of each object type do you see? Does the count exactly match what was requested? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_Counting_Hard | Spatial Reasoning | Counting | Hard | Exactly 9 goldfish swimming in a large round fishbowl | A fruit bowl containing exactly 3 bananas, 4 oranges, and 2 apples | Exactly 11 colorful balloons floating against a clear blue sky | How many goldfish are in the image? PASS if there are exactly 9; FAIL for any other number. | Count the fruit: how many bananas, oranges, and apples are there? PASS only if there are exactly 3 bananas, 4 oranges, and 2 apples; FAIL if any count differs. | How many balloons are in the image? PASS if there are exactly 11; FAIL for any other number. | Exact count match | Count the objects in this image. The prompt asked for: "{prompt_excerpt}". How many of each object type do you see? Does the count exactly match what was requested? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_AttributesBinding_Easy | Spatial Reasoning | Attributes Binding | Easy | A red ball and a blue cube sitting side by side on a white surface | A green apple and an orange pumpkin on a wooden cutting board | A yellow taxi and a black limousine parked next to each other on a street | Is there a RED ball and a BLUE cube, each with the correct colour? PASS only if the ball is red AND the cube is blue; FAIL if either colour is wrong or swapped. | Is there a GREEN apple and an ORANGE pumpkin, each with the correct colour? PASS only if the apple is green AND the pumpkin is orange; FAIL if either colour is wrong or swapped. | Is there a YELLOW taxi and a BLACK limousine, each with the correct colour? PASS only if the taxi is yellow AND the limousine is black; FAIL if either colour is wrong or swapped. | Each color on correct object — no swap | The prompt asked for specific objects with specific attributes: "{prompt_excerpt}". Is each attribute (color, size, shape) correctly bound to the right object with no swaps? List each object and its observed attributes. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_AttributesBinding_Medium | Spatial Reasoning | Attributes Binding | Medium | A large red circle, a small blue triangle, and a medium green star arranged on a white canvas | A tall thin white candle, a short fat red candle, and a medium spiral yellow candle on a shelf | A big brown horse, a small white rabbit, and a medium-sized black cat standing in a meadow | Are there a large RED circle, a small BLUE triangle, and a medium GREEN star, each with the correct colour and size? PASS only if all three match (red large circle, blue small triangle, green medium star); FAIL if any colour, size, or shape is wrong or swapped. | Are there a tall thin WHITE candle, a short fat RED candle, and a medium spiral YELLOW candle, each with the correct colour and shape? PASS only if all three match; FAIL if any colour, size, or shape is wrong or swapped. | Are there a big BROWN horse, a small WHITE rabbit, and a medium-sized BLACK cat, each with the correct colour and size? PASS only if all three match; FAIL if any colour or size is wrong or swapped. | All attributes correctly bound | The prompt asked for specific objects with specific attributes: "{prompt_excerpt}". Is each attribute (color, size, shape) correctly bound to the right object with no swaps? List each object and its observed attributes. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_AttributesBinding_Hard | Spatial Reasoning | Attributes Binding | Hard | A pink fire truck, a blue banana, and a green sun in a surreal landscape | A purple carrot, a red tree with red leaves, and a white watermelon sliced open | An orange elephant, a yellow penguin, and a silver strawberry arranged on a table | Are there a PINK fire truck, a BLUE banana, and a GREEN sun, each with these exact colours? PASS only if the fire truck is pink AND the banana is blue AND the sun is green; FAIL if any colour is wrong, default, or swapped. | Are there a PURPLE carrot, a RED tree with red leaves, and a WHITE watermelon sliced open, each with these exact colours? PASS only if the carrot is purple AND the tree/leaves are red AND the watermelon is white; FAIL if any colour is wrong, default, or swapped. | Are there an ORANGE elephant, a YELLOW penguin, and a SILVER strawberry, each with these exact colours? PASS only if the elephant is orange AND the penguin is yellow AND the strawberry is silver; FAIL if any colour is wrong, default, or swapped. | Non-default attributes respected | The prompt asked for specific objects with specific attributes: "{prompt_excerpt}". Is each attribute (color, size, shape) correctly bound to the right object with no swaps? List each object and its observed attributes. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_RelativePosition_Easy | Spatial Reasoning | Relative Position | Easy | A cat sitting on top of a cardboard box | A red ball underneath a wooden chair | A coffee mug next to an open book on a desk | Is the cat positioned on top of the cardboard box? PASS if the cat is clearly sitting on top of the box; FAIL otherwise. | Is the red ball positioned underneath the wooden chair? PASS if the ball is clearly under the chair; FAIL otherwise. | Is the coffee mug positioned next to the open book? PASS if the mug is clearly beside the open book; FAIL otherwise. | Spatial relation correct — binary | The prompt specified spatial relationships: "{prompt_excerpt}". Describe the spatial arrangement of the objects. Are all described positional relationships (on, under, left, right, above, below, etc..) correct? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_RelativePosition_Medium | Spatial Reasoning | Relative Position | Medium | A bird perched on a branch, the branch extending from a tree next to a park bench | A cat sitting on a pillow, the pillow placed on a wooden chair | A candle on top of a stack of books, the books next to a potted plant | Is a bird perched on a branch that extends from a tree standing next to a park bench? PASS only if the bird is on the branch AND the tree is next to a bench; FAIL if any relation is missing or wrong. | Is a cat sitting on a pillow that is placed on a wooden chair? PASS only if the cat is on the pillow AND the pillow is on the chair; FAIL if any relation is missing or wrong. | Is a candle on top of a stack of books, with the books next to a potted plant? PASS only if the candle is on the books AND the books are beside a potted plant; FAIL if any relation is missing or wrong. | All relations correct | The prompt specified spatial relationships: "{prompt_excerpt}". Describe the spatial arrangement of the objects. Are all described positional relationships (on, under, left, right, above, below, etc..) correct? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_RelativePosition_Hard | Spatial Reasoning | Relative Position | Hard | Five chess pieces on a board: a king in the center, a queen to its left, a bishop to its right, a rook above the king, and a knight below the king | A desk scene: laptop in the center, coffee mug to the left, phone to the right, lamp above, and notebook below | Five animals in a field: a horse on the far left, a cow to the right of the horse, a sheep in the center, a goat to the right of the sheep, and a dog on the far right | Among the chess pieces, is the king in the center with the queen to its left, the bishop to its right, the rook above it, and the knight below it? PASS only if all five positions are correct; FAIL if any is wrong. | On the desk, is the laptop in the center with the coffee mug to its left, the phone to its right, the lamp above it, and the notebook below it? PASS only if all five positions are correct; FAIL if any is wrong. | From left to right, are the animals ordered horse, cow, sheep (center), goat, then dog on the far right? PASS only if the left-to-right order is horse, cow, sheep, goat, dog; FAIL if any is out of order. | All positions match prompt | The prompt specified spatial relationships: "{prompt_excerpt}". Describe the spatial arrangement of the objects. Are all described positional relationships (on, under, left, right, above, below, etc..) correct? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_RelativePosition_Extreme | Spatial Reasoning | Relative Position | Extreme | A landscape with a red barn in the foreground at large scale, a windmill in the midground at medium scale, and snow-capped mountains in the background at small scale, all with consistent depth perspective | A beach scene with a sandcastle in the foreground, a surfer on the waves in the midground, and a cruise ship on the horizon in the background, sizes decreasing with distance | A city park with a park bench and pigeons in the foreground, a fountain in the midground, and skyscrapers in the background, each layer progressively smaller | Is a large red barn in the foreground, a medium windmill in the midground, and small snow-capped mountains in the background, with sizes decreasing by depth? PASS only if all three depth layers are present and their sizes shrink with distance; FAIL otherwise. | Is a sandcastle in the foreground, a surfer on the waves in the midground, and a cruise ship on the horizon in the background, with sizes decreasing by distance? PASS only if all three depth layers are present and their sizes shrink with distance; FAIL otherwise. | Is a park bench with pigeons in the foreground, a fountain in the midground, and skyscrapers in the background, with each layer progressively smaller? PASS only if all three depth layers are present and their sizes shrink with distance; FAIL otherwise. | Depth ordering + scale coherence | The prompt specified spatial relationships: "{prompt_excerpt}". Describe the spatial arrangement of the objects. Are all described positional relationships (on, under, left, right, above, below, etc..) correct? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_Negation_Easy | Spatial Reasoning | Negation | Easy | A tropical beach scene with no people | A cat lying on a couch without a collar | A classroom with desks and a chalkboard but no students | Are any people visible anywhere in this beach scene? PASS if no people are visible; FAIL if any person appears. | Is the cat wearing a collar? PASS if the cat has no collar; FAIL if a collar is visible. | Are any students visible in this classroom? PASS if no students are present; FAIL if any student appears. | Excluded element truly absent — binary | The prompt explicitly excluded certain elements: "{prompt_excerpt}". Carefully check: is the excluded element truly absent from the image? Describe what you see. Rate: PASS if excluded element is absent, FAIL if present. | google/gemini-3.1-pro-preview |
SpatialReasoning_Negation_Medium | Spatial Reasoning | Negation | Medium | A sandy beach with palm trees, seashells, and a beach umbrella but no water or ocean visible | An office workspace with a desk, chair, bookshelf, and lamp but no computers or screens | A kitchen with pots, pans, a stove, and a cutting board but no food | Is any water or ocean visible anywhere in the image? PASS if no water or ocean is visible at all; FAIL if any water or ocean appears. | Are any computers, monitors, or screens visible in the image? PASS if none are present; FAIL if any computer or screen appears. | Is any food visible in the image? PASS if no food is present; FAIL if any food appears. | Excluded element absent despite strong prior | The prompt explicitly excluded certain elements: "{prompt_excerpt}". Carefully check: is the excluded element truly absent from the image? Describe what you see. Rate: PASS if excluded element is absent, FAIL if present. | google/gemini-3.1-pro-preview |
SpatialReasoning_Negation_Hard | Spatial Reasoning | Negation | Hard | A pepperoni pizza with tomato sauce, basil, and crust but absolutely no cheese | A car body with windows, doors, and headlights but no wheels | A human face with eyes, nose, and mouth but no eyebrows | Is there any cheese on this pizza? PASS only if the pizza has tomato sauce, basil, and crust but NO cheese; FAIL if any cheese is visible or if the required elements are missing. | Does this car have any wheels? PASS only if the car body has windows, doors, and headlights but NO wheels; FAIL if any wheel is visible or if the required elements are missing. | Does this face have eyebrows? PASS only if the face has eyes, nose, and mouth but NO eyebrows; FAIL if any eyebrow is visible or if the required features are missing. | Correct inclusion AND exclusion | The prompt explicitly excluded certain elements: "{prompt_excerpt}". Carefully check: is the excluded element truly absent from the image? Describe what you see. Rate: PASS if excluded element is absent, FAIL if present. | google/gemini-3.1-pro-preview |
SpatialReasoning_ScaleProportions_Easy | Spatial Reasoning | Scale & Proportions | Easy | A person standing next to a yellow school bus, showing realistic relative scale | A ladybug sitting on a large green leaf, shown at realistic proportions | A child holding a basketball, both at correct relative sizes | Is the person shown at a realistic size relative to the yellow school bus (much smaller than the bus)? PASS if the relative scale is realistic; FAIL if proportions are clearly wrong. | Is the ladybug shown much smaller than the green leaf, at realistic proportions? PASS if the ladybug is realistically tiny relative to the leaf; FAIL if proportions are clearly wrong. | Is the basketball shown at a realistic size relative to the child (roughly the size the child can hold)? PASS if the relative scale is realistic; FAIL if proportions are clearly wrong. | Relative sizes correct | The prompt described specific size relationships: "{prompt_excerpt}". Think about the relative size of what is described in the prompt, and look at the image. Are the relative sizes/proportions of the objects correct and consistent? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_ScaleProportions_Medium | Spatial Reasoning | Scale & Proportions | Medium | A tiny person standing on the rim of a coffee cup, looking down into it | A giant tabby cat walking between city skyscrapers like a kaiju | A miniature house sitting on the palm of someone's hand | Does the image show a tiny person standing on the rim of a coffee cup, with the person dramatically smaller than the cup? PASS if the person is clearly miniature relative to the cup; FAIL otherwise. | Does the image show a giant cat towering among city skyscrapers at kaiju scale, with the cat as large as or larger than the buildings? PASS if the cat is clearly giant relative to the skyscrapers; FAIL otherwise. | Does the image show a miniature house resting on the palm of a hand, with the house tiny relative to the hand? PASS if the house is clearly miniature relative to the hand; FAIL otherwise. | Scale contrast clearly intentional + visually coherent | The prompt described specific size relationships: "{prompt_excerpt}". Think about the relative size of what is described in the prompt, and look at the image. Are the relative sizes/proportions of the objects correct and consistent? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_ScaleProportions_Hard | Spatial Reasoning | Scale & Proportions | Hard | A group photo of an elephant, a horse, a dog, a cat, and a mouse, all standing in a line with correct real-world proportional sizes | A dinner table scene with a plate, fork, knife, wine glass, salt shaker, and peppercorn, all at realistic relative sizes | A farm scene with a barn, tractor, farmer, cow, chicken, and a fence post, all at correct proportional scale | Are an elephant, horse, dog, cat, and mouse lined up with correct real-world relative sizes (elephant largest, then horse, dog, cat, mouse smallest)? PASS only if the size ordering is realistic across all five; FAIL if any is disproportionate. | Are a plate, fork, knife, wine glass, salt shaker, and peppercorn all shown at realistic relative sizes (e.g., peppercorn tiny, plate largest)? PASS only if all proportions match real-world scale; FAIL if any is disproportionate. | Are a barn, tractor, farmer, cow, chicken, and fence post all shown at correct real-world relative sizes (barn largest, chicken small, etc.)? PASS only if all proportions match real-world scale; FAIL if any is disproportionate. | All proportions consistent with real-world | The prompt described specific size relationships: "{prompt_excerpt}". Think about the relative size of what is described in the prompt, and look at the image. Are the relative sizes/proportions of the objects correct and consistent? Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_Compositionality_Easy | Spatial Reasoning | Compositionality | Easy | A tabby cat wearing oversized red sunglasses | A golden retriever wearing a tiny cowboy hat | A penguin holding a colorful umbrella | Is there a tabby cat wearing oversized red sunglasses? PASS only if a tabby cat AND oversized red sunglasses worn by it are both present; FAIL if either is missing. | Is there a golden retriever wearing a tiny cowboy hat? PASS only if a golden retriever AND a tiny cowboy hat worn by it are both present; FAIL if either is missing. | Is there a penguin holding a colorful umbrella? PASS only if a penguin AND a colorful umbrella held by it are both present; FAIL if either is missing. | Both concepts present and combined | The prompt combined unusual concepts: "{prompt_excerpt}". Read the prompt and look at the image. Are ALL described elements present and coherently combined in the image? List each element and whether it's present. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_Compositionality_Medium | Spatial Reasoning | Compositionality | Medium | A medieval knight in full armor sitting at a desk typing on a modern laptop | A Roman gladiator taking a selfie with a smartphone in the Colosseum | An astronaut in a spacesuit riding a bicycle through a park | Is there a medieval knight in full armor sitting at a desk and typing on a modern laptop? PASS only if the armored knight AND the modern laptop are both present and combined; FAIL if either element is missing. | Is there a Roman gladiator taking a selfie with a smartphone in the Colosseum? PASS only if the gladiator, the smartphone selfie, and the Colosseum setting are all present; FAIL if any element is missing. | Is there an astronaut in a spacesuit riding a bicycle through a park? PASS only if the spacesuited astronaut AND the bicycle in a park setting are both present and combined; FAIL if any element is missing. | Both domains convincingly merged | The prompt combined unusual concepts: "{prompt_excerpt}". Read the prompt and look at the image. Are ALL described elements present and coherently combined in the image? List each element and whether it's present. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
SpatialReasoning_Compositionality_Hard | Spatial Reasoning | Compositionality | Hard | An octopus wearing a top hat and monocle playing chess on the surface of the moon with Earth visible in the background | A dinosaur in a business suit giving a PowerPoint presentation in a modern office while drinking coffee | A robot dressed as a pirate sailing a paper boat on a lake of lava while juggling three kittens | Is there an octopus wearing a top hat and monocle playing chess on the moon's surface with Earth in the background? PASS only if the octopus, top hat, monocle, chess game, lunar surface, and Earth in the background are ALL present; FAIL if any element is missing. | Is there a dinosaur in a business suit giving a PowerPoint presentation in a modern office while drinking coffee? PASS only if the suited dinosaur, the PowerPoint presentation, the office setting, and the coffee are ALL present; FAIL if any element is missing. | Is there a robot dressed as a pirate sailing a paper boat on a lake of lava while juggling three kittens? PASS only if the pirate robot, the paper boat, the lava lake, and three juggled kittens are ALL present; FAIL if any element is missing. | All elements present and coherent | The prompt combined unusual concepts: "{prompt_excerpt}". Read the prompt and look at the image. Are ALL described elements present and coherently combined in the image? List each element and whether it's present. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_Hands_Easy | Human realism | Hands | Easy | A person standing with both hands visible at their sides, fingers relaxed and naturally spread | A woman holding her hands out in front of her palms-up, all ten fingers clearly visible | A man with arms crossed, both hands resting on opposite upper arms with fingers visible | Look at the person's hands hanging at their sides. Is each visible hand anatomically normal with exactly five naturally-spread fingers and correct proportions (no extra, missing, or fused digits, no distortions)? PASS only if every visible hand is anatomically correct; FAIL otherwise. | Look at the woman holding both hands out palms-up. Are all ten fingers clearly visible and are both hands anatomically normal (exactly five fingers each, natural proportions, no extra, missing, or fused digits)? PASS only if both hands are correct; FAIL otherwise. | Look at the person with arms crossed, each hand resting on the opposite upper arm. Is each visible hand anatomically normal with five fingers and correct proportions (no extra, missing, or fused digits, no distortions)? PASS only if the hands are anatomically correct; FAIL otherwise. | Correct finger count (5 per hand), natural proportions | Examine the hands in this image closely. The prompt was: "{prompt_excerpt}". Is the hand or the hands looks normal? Count the fingers to do an extra check. on each visible hand. Are there exactly 5 per hand? Is the grip/gesture anatomically correct? Note any anomalies. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_Hands_Medium | Human realism | Hands | Medium | A person holding a white ceramic coffee mug by the handle with their right hand | A woman writing in a notebook with a pen gripped correctly in her right hand | A person playing an acoustic guitar with left hand on the fretboard and right hand strumming | Look at the hand holding a white ceramic coffee mug by its handle. Is the grip plausible and is the hand anatomically normal with exactly five fingers (no extra, missing, or fused digits, no distortions)? PASS only if the mug is properly held AND the hand is correct; FAIL otherwise. | Look at the woman writing in a notebook with a pen. Is the pen gripped in a natural, correct writing hold and is the hand anatomically normal with exactly five fingers (no extra, missing, or fused digits)? PASS only if the grip is correct AND the hand is anatomically normal; FAIL otherwise. | Look at the person playing an acoustic guitar, with one hand on the fretboard and the other strumming. Are both hands positioned plausibly on the guitar and anatomically normal (five fingers each, no extra, missing, or fused digits, no distortions)? PASS only if both hands are correct and plausibly placed; FAIL otherwi... | Correct grip + correct finger count + object properly held | Examine the hands in this image closely. The prompt was: "{prompt_excerpt}". Count the fingers on each visible hand. Are there exactly 5 per hand? Is the grip/gesture anatomically correct? Note any anomalies. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_Hands_Hard | Human realism | Hands | Hard | A person making a peace sign with their right hand, clearly showing two fingers up and three curled | A hand giving a thumbs-up with four fingers curled and thumb extended upward | A person counting to three on their fingers, with index, middle, and ring fingers extended | Look at the hand. Is it clearly making a peace sign — exactly two fingers extended and the rest curled — with five normal fingers and no anatomical defects (no extra, missing, or fused digits)? PASS only if the gesture is a correct peace sign and the hand looks anatomically normal; FAIL otherwise. | Is the hand clearly a thumbs-up — thumb extended upward, four fingers curled — with normal anatomy (five fingers, no extra/missing/fused digits)? PASS if yes; FAIL otherwise. | Is the hand showing exactly three fingers extended (index, middle, ring) with normal anatomy and no digit defects? PASS if three fingers are up and the hand is anatomically correct; FAIL otherwise. | Gesture accurate + anatomy correct | Examine the hands in this image closely. The prompt was: "{prompt_excerpt}". Count the fingers on each visible hand. Are there exactly 5 per hand? Is the grip/gesture anatomically correct? Note any anomalies. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_Hands_Extreme | Human realism | Hands | Extreme | Two people shaking hands in a professional setting, all fingers of both hands anatomically correct | Two people doing a high-five, palms meeting with all fingers extended on both hands | One person handing a small gift box to another person, both sets of hands gripping the box correctly | Are exactly two people shown shaking hands, with the clasped hands anatomically normal (five fingers each, natural interlocking grip, no extra, missing, or fused digits or distortions)? PASS only if it clearly reads as a handshake AND all visible hands are correct; FAIL otherwise. | Are exactly two people shown doing a high-five with palms meeting and fingers extended, and are both hands anatomically normal (five fingers each, no extra, missing, or fused digits or distortions)? PASS only if it clearly reads as a high-five AND both hands are correct; FAIL otherwise. | Is one person handing a small gift box to another, with both people's hands gripping the box plausibly and anatomically normal (five fingers each, no extra, missing, or fused digits or distortions)? PASS only if the box-handoff reads clearly AND all visible hands are correct; FAIL otherwise. | All hands anatomically correct + interaction coherent | Examine the hands in this image closely. The prompt was: "{prompt_excerpt}". Count the fingers on each visible hand. Are there exactly 5 per hand? Is the grip/gesture anatomically correct? Note any anomalies. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FacesExpressions_Easy | Human realism | Faces & Expressions | Easy | A close-up portrait of a young woman with a warm natural smile | A portrait of a middle-aged man with a calm, serious expression | A headshot of a teenager with a relaxed, neutral face | Look at the close-up portrait of a young woman. Does she show a warm, natural smile and is her face anatomically normal and symmetric with correct features and no distortions? PASS only if the warm smile reads clearly AND the face is undistorted; FAIL otherwise. | Look at the portrait of a middle-aged man. Does his expression read as calm and serious, and is his face anatomically normal and symmetric with correct features and no distortions? PASS only if the calm/serious expression reads AND the face is undistorted; FAIL otherwise. | Look at the headshot of a teenager. Does the face show a relaxed, neutral expression and look anatomically normal and symmetric with correct features and no distortions? PASS only if the neutral expression reads AND the face is undistorted; FAIL otherwise. | Symmetric face, correct features, recognizable emotion | Look at the face(s) in this image. The prompt asked for: "{prompt_excerpt}". Is the face symmetric with correct features? Is the intended emotion clearly readable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FacesExpressions_Medium | Human realism | Faces & Expressions | Medium | A close-up of a person's face showing clear disgust, nose wrinkled and lip curled | A portrait capturing genuine surprise — raised eyebrows, wide eyes, slightly open mouth | A face showing quiet melancholy — downcast eyes, slightly drawn mouth, distant gaze | Does the face clearly show disgust (e.g. wrinkled nose, curled lip) and look anatomically normal with no distortions? PASS if the disgust expression is clearly readable and the face is undistorted; FAIL otherwise. | Does the face clearly show surprise (raised eyebrows, wide eyes, slightly open mouth) with normal, undistorted features? PASS if the surprise reads clearly and the face is undistorted; FAIL otherwise. | Does the face clearly convey quiet sadness / melancholy (downcast eyes, drawn mouth) with normal, undistorted features? PASS if the melancholy reads clearly and the face is undistorted; FAIL otherwise. | Emotion clearly readable and convincing | Look at the face(s) in this image. The prompt asked for: "{prompt_excerpt}". Is the face symmetric with correct features? Is the intended emotion clearly readable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FacesExpressions_Hard | Human realism | Faces & Expressions | Hard | An elderly woman in her 80s with deep wrinkles and white hair, smiling warmly with kind eyes | A 5-year-old child with round cheeks and large eyes, crying with tears streaming down their face | An elderly man in his 70s with age spots and thinning gray hair, looking skeptically over reading glasses | Look at the elderly woman with white hair. Does she appear to be in her 80s with deep wrinkles, smiling warmly with kind eyes, and is her face anatomically normal with no distortions? PASS only if the age reads as elderly, the warm smile is clear, AND the face is undistorted; FAIL otherwise. | Look at the young child (around 5, round cheeks, large eyes). Is the child clearly crying with visible tears streaming down the face, and is the face anatomically normal with no distortions? PASS only if the crying with tears reads clearly, the age reads as a young child, AND the face is undistorted; FAIL otherwise. | Look at the elderly man with thinning gray hair and reading glasses. Does he appear to be in his 70s looking skeptically over the glasses, and is his face anatomically normal with no distortions? PASS only if the age reads as elderly, the skeptical look reads clearly, AND the face is undistorted; FAIL otherwise. | Age-appropriate features + emotion match | Look at the face(s) in this image. The prompt asked for: "{prompt_excerpt}". Is the face symmetric with correct features? Is the intended emotion clearly readable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FacesExpressions_Extreme | Human realism | Faces & Expressions | Extreme | A group of three friends at a table: one laughing with head tilted back, one looking shocked with wide eyes, and one rolling their eyes with a smirk | A family photo with four people: a father beaming with pride, a mother smiling gently, a teenager looking bored, and a child grinning excitedly | Three coworkers in a meeting: one looking frustrated with furrowed brows, one nodding with a pleased expression, and one looking confused with a tilted head | Are there exactly three people at a table showing three distinct expressions — one laughing with head tilted back, one shocked with wide eyes, and one rolling their eyes with a smirk — each face anatomically normal and undistorted? PASS only if all three people are present with those distinct, readable expressions AND ... | Are there exactly four people showing four distinct expressions — a proud beaming father, a gently smiling mother, a bored teenager, and an excitedly grinning child — each face anatomically normal and undistorted? PASS only if all four are present with those distinct, readable expressions AND normal faces; FAIL otherwi... | Are there exactly three coworkers showing three distinct expressions — one frustrated with furrowed brows, one nodding with a pleased expression, and one confused with a tilted head — each face anatomically normal and undistorted? PASS only if all three are present with those distinct, readable expressions AND normal f... | Each face distinct + each expression different and readable | Look at the face(s) in this image. The prompt asked for: "{prompt_excerpt}". Is the face symmetric with correct features? Is the intended emotion clearly readable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FullBody_Easy | Human realism | Full Body | Easy | A woman standing upright in a relaxed pose, arms at her sides, wearing jeans and a white t-shirt | A man standing straight facing the camera, hands in pockets, normal proportions | A person standing in a neutral pose in a park, full body visible from head to feet | Look at the woman standing in a relaxed pose with arms at her sides, wearing jeans and a white t-shirt. Is her body anatomically correct with normal proportions and the right number of limbs (no extra, missing, or distorted limbs)? PASS only if the pose matches and the body is anatomically normal; FAIL otherwise. | Look at the man standing straight and facing the camera with his hands in his pockets. Is his body anatomically correct with normal proportions and the right number of limbs (no extra, missing, or distorted limbs)? PASS only if the pose matches and the body is anatomically normal; FAIL otherwise. | Look at the person standing in a neutral pose, full body visible from head to feet. Is the body anatomically correct with normal proportions and the right number of limbs (no extra, missing, or distorted limbs)? PASS only if the full body is visible and anatomically normal; FAIL otherwise. | Correct limb count, natural proportions, no distortion | Examine the full body in this image. The prompt was: "{prompt_excerpt}". Are limb count, proportions, and pose anatomically correct? Is the described action/pose recognizable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FullBody_Medium | Human realism | Full Body | Medium | A woman mid-stride running through a park, hair and clothes showing motion | A man captured mid-jump over a puddle, body in the air with bent knees | A dancer in a dynamic spin, dress flowing outward with one leg extended | Look at the woman shown mid-stride running, with hair and clothes conveying motion. Is the running pose anatomically plausible with correct proportions and the right number of limbs (no extra, missing, fused, or distorted limbs)? PASS only if the running action reads clearly AND the body is anatomically correct; FAIL o... | Look at the man captured mid-jump over a puddle, airborne with bent knees. Is the jumping pose anatomically plausible with correct proportions and the right number of limbs (no extra, missing, fused, or distorted limbs)? PASS only if the mid-jump action reads clearly AND the body is anatomically correct; FAIL otherwise... | Look at the dancer in a dynamic spin with dress flowing outward and one leg extended. Is the spinning pose anatomically plausible with correct proportions and the right number of limbs (no extra, missing, fused, or distorted limbs)? PASS only if the dynamic spin reads clearly AND the body is anatomically correct; FAIL ... | Anatomically plausible pose + correct proportions | Examine the full body in this image. The prompt was: "{prompt_excerpt}". Are limb count, proportions, and pose anatomically correct? Is the described action/pose recognizable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FullBody_Hard | Human realism | Full Body | Hard | A person in a yoga tree pose — standing on one leg with the other foot pressed to the inner thigh, arms raised overhead | A rock climber clinging to a cliff face with arms and legs extended to different holds | A cellist seated with correct posture, legs around the instrument, bow arm extended | Look at the person in a yoga tree pose — standing on one leg with the other foot pressed to the inner thigh and arms raised overhead. Does the pose match that description and is the body anatomically correct with the right number of limbs and no distortions? PASS only if the tree pose is recognizable AND the anatomy is... | Look at the rock climber clinging to a cliff face with arms and legs reaching to different holds. Is the climbing pose plausible and is the body anatomically correct with the right number of limbs (no extra, missing, fused, or distorted limbs)? PASS only if the climbing pose reads clearly AND the anatomy is correct; FA... | Look at the seated cellist with the instrument between the legs and the bow arm extended. Is the posture plausible for playing a cello and is the body anatomically correct with the right number of limbs and no distortions? PASS only if the cello-playing posture reads clearly AND the anatomy is correct; FAIL otherwise. | Pose recognizable + anatomy correct + limbs coherent | Examine the full body in this image. The prompt was: "{prompt_excerpt}". Are limb count, proportions, and pose anatomically correct? Is the described action/pose recognizable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_FullBody_Extreme | Human realism | Full Body | Extreme | Two tango dancers in a close embrace, the man dipping the woman backward with intertwined legs | Two soccer players mid-tackle, one sliding and the other jumping over with the ball at their feet | Two gymnasts performing a synchronized handstand, side by side with identical body alignment | Are there exactly two tango dancers in a close embrace, the man dipping the woman backward with intertwined legs, both bodies anatomically correct (right number of limbs, no fused, extra, or distorted limbs) and the interaction physically plausible? PASS only if two dancers in that dip pose read clearly AND both bodies... | Are there exactly two soccer players mid-tackle — one sliding and the other jumping over with the ball at their feet — both bodies anatomically correct (right number of limbs, no fused, extra, or distorted limbs) and the interaction physically plausible? PASS only if the two-player tackle reads clearly AND both bodies ... | Are there exactly two gymnasts performing a synchronized handstand side by side with matching body alignment, both bodies anatomically correct (right number of limbs, no fused, extra, or distorted limbs)? PASS only if two side-by-side handstands read clearly AND both bodies are correct; FAIL otherwise. | All bodies correct + spatial interaction plausible | Examine the full body in this image. The prompt was: "{prompt_excerpt}". Are limb count, proportions, and pose anatomically correct? Is the described action/pose recognizable? Note any distortions. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_MultiSubject_Easy | Human realism | Multi-Subject | Easy | Two women standing side by side: one with short blonde hair wearing a red jacket, the other with long black hair wearing a blue dress | Two men at a cafe: one with a beard and glasses wearing a green sweater, the other clean-shaven in a black hoodie | Two friends posing: one tall with curly red hair in denim overalls, the other short with straight brown hair in a yellow sundress | Are there exactly two women side by side — one with short blonde hair in a red jacket, the other with long black hair in a blue dress — visually distinct (features not merged) and each anatomically normal? PASS only if both women are present with those matching distinct features AND normal anatomy; FAIL otherwise. | Are there exactly two men at a cafe — one bearded with glasses in a green sweater, the other clean-shaven in a black hoodie — visually distinct (features not merged) and each anatomically normal? PASS only if both men are present with those matching distinct features AND normal anatomy; FAIL otherwise. | Are there exactly two friends — one tall with curly red hair in denim overalls, the other short with straight brown hair in a yellow sundress — visually distinct (features not merged) and each anatomically normal? PASS only if both are present with those matching distinct features AND normal anatomy; FAIL otherwise. | Two separate individuals, features not merged | The prompt described multiple distinct people: "{prompt_excerpt}". Are all individuals present, visually distinct, and matching their described features? List each person and their observed traits. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
HumanRealism_MultiSubject_Medium | Human realism | Multi-Subject | Medium | Four people at a dinner table: a bald man in a suit, a woman with silver hair in a red blouse, a young man with dreadlocks in a denim jacket, and a woman with a hijab in a green dress | Three musicians on stage: a tall woman with short pink hair playing bass, a man with a beard and bandana on drums, and a slim man with glasses playing keyboard | Three hikers on a trail: a man with a backpack and sunhat, a woman with braided hair and hiking poles, and a teenager with a baseball cap and water bottle | Are there exactly four distinct people at a dinner table — a bald man in a suit, a silver-haired woman in a red blouse, a young man with dreadlocks in a denim jacket, and a woman in a hijab in a green dress — each matching those features and anatomically normal? PASS only if all four are present with their distinct mat... | Are there exactly three distinct musicians on stage — a tall woman with short pink hair on bass, a bearded man with a bandana on drums, and a slim man with glasses on keyboard — each matching those features and anatomically normal? PASS only if all three are present with their distinct matching features AND normal anat... | Are there exactly three distinct hikers on a trail — a man with a backpack and sunhat, a woman with braided hair and hiking poles, and a teenager with a baseball cap and water bottle — each matching those features and anatomically normal? PASS only if all three are present with their distinct matching features AND norm... | All individuals distinct and identifiable | The prompt described multiple distinct people: "{prompt_excerpt}". Are all individuals present, visually distinct, and matching their described features? List each person and their observed traits. Rate: PASS or FAIL. | google/gemini-3.1-pro-preview |
Truthfulness_Photorealism_Medium | Truthfulness | Photorealism | Medium | A still life on a wooden table featuring a clear glass of water, a polished silver spoon, a folded linen napkin, and a ripe red apple — each material rendered with accurate texture and light behavior | A workspace desk with a brushed aluminum laptop, a leather-bound notebook, a ceramic coffee mug, and a glass paperweight — all materials visually distinct and realistic | A kitchen counter with a stainless steel pot, a wooden cutting board, a glass bottle of olive oil, and a cotton dish towel — each surface reflecting light correctly | Does the image look like a real photograph of a still life on a wooden table with a clear glass of water, a polished silver spoon, a folded linen napkin, and a ripe red apple, each material rendered with convincing texture and light behavior (glass translucency, metallic reflections, fabric weave, waxy apple skin)? PAS... | Does the image look like a real photograph of a workspace desk with a brushed aluminum laptop, a leather-bound notebook, a ceramic coffee mug, and a glass paperweight, with each material visually distinct and realistic (matte brushed metal, grained leather, glazed ceramic, transparent glass)? PASS if it reads as a real... | Does the image look like a real photograph of a kitchen counter with a stainless steel pot, a wooden cutting board, a glass bottle of olive oil, and a cotton dish towel, each surface reflecting light correctly (specular steel highlights, matte wood grain, translucent glass and oil, soft fabric)? PASS if it reads as a r... | Each material renders with correct visual properties | The prompt asked for multiple materials: "{prompt_excerpt}". Does each material (glass, metal, wood, fabric, etc.) render with correct visual properties (reflections, texture, translucency)? Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_PhysicsReflections_Easy | Truthfulness | Physics & Reflections | Easy | A red sphere on a white surface with a single shadow cast to the right, consistent with a light source from the upper left | A coffee mug on a sunny windowsill casting a sharp shadow to the left | A person standing outdoors at midday with a short shadow directly below them | Does the image show a red sphere on a white surface casting a single shadow to the right, consistent with a light source coming from the upper left? PASS if the shadow falls to the right and matches an upper-left light; FAIL if the shadow is missing, points the wrong way, or is inconsistent with the lighting. | Does the image show a coffee mug on a sunny windowsill casting a single sharp shadow to the left? PASS if the mug casts a sharp shadow to the left consistent with sunlight; FAIL if the shadow is missing, blurry when it should be sharp, or cast in the wrong direction. | Does the image show a person standing outdoors at midday with a short shadow directly below them? PASS if the shadow is short and falls beneath the person as expected for a high midday sun; FAIL if the shadow is long, absent, or cast to one side. | Shadow direction + length plausible | The prompt involved shadows/reflections: "{prompt_excerpt}". Are the shadows/reflections physically plausible? Check direction, orientation, distortion, and color. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_PhysicsReflections_Medium | Truthfulness | Physics & Reflections | Medium | A woman standing in front of a full-length mirror, her reflection accurately showing the back of her outfit and reversed orientation | A cat sitting on a polished marble floor with its reflection visible below | A vase of flowers on a glass table with a clear reflection visible on the table surface | Does the full-length mirror show a correct reflection of the woman, revealing the back of her outfit with left-right reversed orientation and matching pose? PASS if the reflection is physically consistent (shows her back and is correctly reversed); FAIL if the reflection shows her front, is missing, or does not match h... | Does the polished marble floor show a correct mirror reflection of the cat directly below it, matching the cat's shape, colour, and pose? PASS if the reflection is present and physically consistent with the cat above; FAIL if it is missing, misaligned, or does not match the cat. | Does the glass table surface show a correct reflection of the vase of flowers, matching its shape and colours in a plausible mirrored position? PASS if the reflection is present and physically consistent with the vase; FAIL if it is missing, distorted incorrectly, or does not match the vase. | Reflection matches object + orientation correct | The prompt involved shadows/reflections: "{prompt_excerpt}". Are the shadows/reflections physically plausible? Check direction, orientation, distortion, and color. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_PhysicsReflections_Hard | Truthfulness | Physics & Reflections | Hard | A row of trees reflected in a calm lake at sunset, with the reflection slightly darker and gently rippled | A city skyline reflected in a rain puddle on a street, with mild distortion and color shift | A mountain range reflected in a still alpine lake, with the reflection inverted and subtly muted in color | Does the calm lake show a reflection of the row of trees at sunset that is inverted, slightly darker than the trees, and gently rippled? PASS if the reflection is present, correctly mirrored, and appropriately darker with mild ripple distortion; FAIL if it is missing, unrippled like a perfect mirror where ripples are e... | Does the rain puddle show a reflection of the city skyline with mild distortion and a slight colour shift? PASS if the skyline reflection is present, plausibly distorted by the puddle, and correctly oriented; FAIL if it is missing, undistorted, or inconsistent with the skyline above. | Does the still alpine lake show a reflection of the mountain range that is inverted and subtly muted in colour compared with the mountains themselves? PASS if the reflection is present, correctly mirrored, and slightly muted; FAIL if it is missing, wrongly oriented, or does not match the mountains. | Reflection present + appropriately distorted + color shift | The prompt involved shadows/reflections: "{prompt_excerpt}". Are the shadows/reflections physically plausible? Check direction, orientation, distortion, and color. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_PhysicsReflections_Extreme | Truthfulness | Physics & Reflections | Extreme | A clear glass sphere sitting on a mirror surface near a pool of water, showing refraction through the sphere, reflection on the mirror, and water reflection simultaneously | A chrome sculpture in a room with a glass floor and large windows, showing multiple reflections and environmental mapping on every surface | A crystal wine glass on a polished marble counter next to a window overlooking a lake — showing refraction in the glass, reflection on the marble, and the lake's own reflective surface | Does the image show a clear glass sphere on a mirror surface near a pool of water, simultaneously displaying refraction (inverted/bent view) through the sphere, a reflection on the mirror beneath it, and a reflection in the water? PASS if all three optical effects are present and physically plausible; FAIL if any of th... | Does the image show a chrome sculpture in a room with a glass floor and large windows, with the chrome reflecting the surrounding environment and further reflections on the glass floor? PASS if the chrome shows plausible environmental reflections and the glass floor reflects the scene consistently; FAIL if reflections ... | Does the image show a crystal wine glass on a polished marble counter beside a window overlooking a lake, displaying refraction through the glass, a reflection on the marble, and the lake's own reflective surface? PASS if all three optical effects are present and physically plausible; FAIL if any of the glass refractio... | All optical effects physically plausible | The prompt involved shadows/reflections: "{prompt_excerpt}". Are the shadows/reflections physically plausible? Check direction, orientation, distortion, and color. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_WorldKnowledge_Easy | Truthfulness | World Knowledge | Easy | The Eiffel Tower in Paris on a clear day, architecturally accurate with its iron lattice structure and three levels | The Taj Mahal in Agra seen from the reflecting pool, with accurate dome, minarets, and symmetry | The Statue of Liberty on her island pedestal, with correct torch, crown, tablet, and green patina | Does the image show the Eiffel Tower in Paris — a wrought-iron lattice tower that tapers upward with three distinct levels and four curved base legs meeting at arches? PASS if it is recognisably the Eiffel Tower with accurate iron-lattice structure; FAIL if the shape, structure, or proportions are wrong or it is a diff... | Does the image show the Taj Mahal in Agra seen from its reflecting pool, with the correct large central onion dome, four flanking minarets, and symmetrical white marble façade? PASS if it is recognisably the Taj Mahal with accurate dome, minarets, and symmetry; FAIL if the structure is inaccurate or it is a different b... | Does the image show the Statue of Liberty on her pedestal with a raised torch in one hand, a spiked seven-point crown, a tablet held in the other arm, and green patina? PASS if it is recognisably the Statue of Liberty with correct torch, crown, tablet, and colour; FAIL if any of these key features are wrong or missing. | Landmark recognizable and architecturally accurate | The prompt asked for a real-world subject: "{prompt_excerpt}". Is the depicted subject recognizable and factually/architecturally accurate? Note any inaccuracies. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_WorldKnowledge_Medium | Truthfulness | World Knowledge | Medium | A traditional Japanese shamisen instrument with accurate three-stringed design, wooden body, and bachi plectrum | A medieval astrolabe with accurate rete, tympan, and engraved degree markings on brass | A Japanese kintsugi bowl — cracked ceramic repaired with visible gold lacquer lines | Does the image show a Japanese shamisen — a three-stringed instrument with a wooden body and a bachi plectrum? PASS if it is recognizably a shamisen with three strings; FAIL if the string count is wrong or it is a different instrument. | Does the image show a medieval brass astrolabe with a rete (star pointers), a tympan/plate, and engraved degree markings? PASS if it is a recognizable astrolabe with these engraved brass details; FAIL otherwise. | Does the image show a ceramic bowl repaired with visible gold lines along the cracks (kintsugi)? PASS if the cracks are filled with gold-coloured lacquer; FAIL if there is no gold-repaired crack. | Object identifiable and details correct | The prompt asked for a real-world subject: "{prompt_excerpt}". Is the depicted subject recognizable and factually/architecturally accurate? Note any inaccuracies. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_WorldKnowledge_Hard | Truthfulness | World Knowledge | Hard | A detailed diagram of the water cycle showing evaporation from the ocean, condensation forming clouds, precipitation as rain, and collection in rivers flowing back to the sea | A cross-section showing tectonic plates at a convergent boundary with subduction, a trench, volcanic arc, and layered mantle | An anatomical illustration of the human heart showing four chambers, valves, aorta, pulmonary arteries, and veins with correct placement | Does the image show an accurate water-cycle diagram depicting evaporation rising from the ocean, condensation forming clouds, precipitation falling as rain, and collection in rivers flowing back to the sea? PASS if all four stages are present and arranged in a scientifically correct cycle; FAIL if a stage is missing, m... | Does the image show a geological cross-section of a convergent plate boundary with one plate subducting beneath another, an oceanic trench, a volcanic arc, and layered mantle? PASS if the subduction, trench, volcanic arc, and layering are all present and geologically correct with all text readable; FAIL if any element ... | Does the image show an anatomically correct human heart with four chambers, valves, the aorta, pulmonary arteries, and veins in correct anatomical placement? PASS if the chambers and major vessels are present and correctly positioned and all text readable. FAIL if chambers are the wrong number, vessels are misplaced, o... | Concept visually accurate and educationally correct | The prompt asked for a real-world subject: "{prompt_excerpt}". Is the depicted subject recognizable and factually/architecturally accurate? Note any inaccuracies. Rate: PASS or FAIL. | qwen36-27b |
Truthfulness_WorldKnowledge_Extreme | Truthfulness | World Knowledge | Extreme | A Morpho menelaus butterfly with accurate iridescent blue dorsal wings, brown ventral wing pattern with eyespots, and correct body morphology | The Hagia Sophia interior showing the massive central dome, pendentives, semi-domes, and Islamic calligraphy medallions alongside Byzantine mosaics | A specimen of bismuth crystal showing accurate iridescent oxide layers, geometric hopper crystal formation, and characteristic staircase structure | Does the image show a Morpho menelaus butterfly with iridescent metallic-blue upper (dorsal) wings, a brown underside patterned with eyespots, and correct butterfly body morphology (six legs, two antennae)? PASS if it is recognisably a Morpho menelaus with accurate iridescent blue wings and correct morphology; FAIL if ... | Does the image show the interior of the Hagia Sophia with its massive central dome resting on pendentives, flanking semi-domes, large Islamic calligraphy medallions, and Byzantine mosaics? PASS if it is recognisably the Hagia Sophia interior with accurate dome, pendentives, medallions, and mosaics; FAIL if the architec... | Does the image show a bismuth crystal specimen with rainbow iridescent oxide surfaces and a geometric stair-stepped hopper crystal structure? PASS if it is recognisably bismuth with accurate iridescent colours and the characteristic staircase/hopper geometry; FAIL if the colour or crystal structure is wrong or it resem... | Subject recognizably correct for an expert | The prompt asked for a real-world subject: "{prompt_excerpt}". Is the depicted subject recognizable and factually/architecturally accurate? Note any inaccuracies. Rate: PASS or FAIL. | qwen36-27b |
ProfessionalStudio_Photorealism_Easy | Professional Studio | Photorealism | Easy | A single red sneaker on a clean white background with soft studio lighting and gentle shadow — product photography style | A clear glass perfume bottle on a reflective black surface with studio lighting, no background distractions | A ripe green apple on a seamless white background with soft side lighting and subtle shadow — catalog style | Does this look like a real studio product photograph of a single red sneaker on a clean white background with soft lighting and a gentle shadow? PASS if it convincingly passes for a real photo of one red sneaker with no obvious AI artifacts; FAIL if there are visible artifacts or the subject/setup is wrong. | Does this look like a real studio product photograph of a clear glass perfume bottle on a reflective black surface, with no distracting background? PASS if it convincingly passes for a real photo of a glass perfume bottle with a clean reflective setup and no obvious AI artifacts; FAIL otherwise. | Does this look like a real catalog-style photograph of a ripe green apple on a seamless white background with soft side lighting and a subtle shadow? PASS if it convincingly passes for a real photo of a green apple with no obvious AI artifacts; FAIL otherwise. | Passes for a photo at first glance — no obvious artifacts | This should be a studio-quality product shot. The prompt was: "{prompt_excerpt}". Does it pass for a real photograph at first glance? Any obvious AI artifacts? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_CameraLighting_Easy | Professional Studio | Camera & Lighting | Easy | A wide-angle shot of a modern living room interior showing expansive space with slight barrel distortion at the edges | A close-up macro shot of a single dewdrop on a rose petal, shallow depth of field | An aerial bird's-eye view looking straight down at a traffic roundabout with cars and trees | Is this a wide-angle interior shot of a modern living room that conveys expansive space, with slight barrel distortion (curved/bowed straight lines) toward the edges? PASS if the wide-angle look and edge distortion are recognisable; FAIL if it looks like a normal/narrow lens with no wide perspective. | Is this a close-up macro shot of a single dewdrop on a rose petal with a shallow depth of field (dewdrop sharp, background softly blurred)? PASS if it reads as a macro image with visible shallow focus; FAIL if everything is in flat focus or the subject is wrong. | Is this an aerial bird's-eye view looking straight down (top-down) at a traffic roundabout with cars and trees? PASS if the viewpoint is clearly directly overhead onto a roundabout; FAIL if the angle is oblique/eye-level or the subject is not a roundabout. | Perspective matches requested camera type | The prompt specified camera/lighting: "{prompt_excerpt}". Does the perspective, depth of field, and lighting pattern match what was described? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_CameraLighting_Medium | Professional Studio | Camera & Lighting | Medium | A portrait lit with classic Rembrandt lighting — triangle of light on the shadowed cheek, moody atmosphere | A landscape photograph bathed in golden hour light — long warm shadows, orange sky, soft glow | A high-key portrait of a person against a bright white background, minimal shadows, airy and luminous | Does the portrait show Rembrandt lighting — a distinct small triangle of light on the shadowed cheek, the rest of the face in moody shadow? PASS if the Rembrandt triangle is recognisable; FAIL if the lighting is flat or lacks the triangle or face distorded | Is the scene lit in golden-hour light — warm orange tones, long soft shadows, a low warm sun? PASS if the warm golden-hour lighting is clearly present; FAIL otherwise. | Is this a high-key image — bright and airy, near-white background, minimal shadows? PASS if the lighting is clearly high-key; FAIL if there are strong shadows or a dark background. | Lighting pattern recognizably matches name | The prompt specified camera/lighting: "{prompt_excerpt}". Does the perspective, depth of field, and lighting pattern match what was described? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_CameraLighting_Hard | Professional Studio | Camera & Lighting | Hard | A portrait shot at 85mm f/1.4 with creamy bokeh background, backlit with warm rim light separating the subject from a blurred garden | A street photograph with shallow depth of field as if shot at 50mm f/1.8, subject sharp with soft city lights in the background | A food photograph shot from above at 35mm f/2.8 with natural window light from the left, slight vignette, warm color temperature | Is this a portrait with a very shallow depth of field (creamy, strongly blurred garden background as from 85mm f/1.4) and a warm rim/backlight separating the subject from the background? PASS if both the smooth bokeh and the warm rim light are present; FAIL if the background is in focus or there is no backlit separatio... | Is this a street photograph with a shallow depth of field where the subject is sharp and the city-light background is softly blurred (as if shot at 50mm f/1.8, with out-of-focus bokeh lights)? PASS if the subject is crisp against a clearly blurred background; FAIL if the whole scene is in focus. | Is this an overhead (top-down) food photograph lit by soft natural window light coming from the left, with a slight vignette and warm color temperature? PASS if the flat-lay overhead angle, directional left-side soft light, and warm tone are all present; FAIL if it is shot at an angle or the lighting direction/warmth i... | DOF + focal length + lighting all match | The prompt specified camera/lighting: "{prompt_excerpt}". Does the perspective, depth of field, and lighting pattern match what was described? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_CameraLighting_Extreme | Professional Studio | Camera & Lighting | Extreme | A tilt-shift photograph of a city street making real-life look like a miniature model — selective focus strip, saturated colors | A cinematic frame with blue and orange anamorphic lens flare streaking horizontally across a dark moody scene | A portrait with dramatic split lighting — half the face brightly lit, the other half in deep shadow, high contrast chiaroscuro | Is this a tilt-shift photograph where a real city street looks like a tiny miniature model — a narrow horizontal strip in sharp focus with the areas above and below strongly blurred, and boosted saturation? PASS if the selective-focus miniature effect is convincing; FAIL if focus is uniform or there is no miniature ill... | Does this dark, moody cinematic frame show a horizontal anamorphic lens flare with distinct blue and orange streaks running across the image? PASS if a horizontal blue/orange lens-flare streak is clearly present; FAIL if there is no horizontal flare or it lacks the blue-orange character. | Is this a portrait with dramatic split lighting — one vertical half of the face brightly lit and the other half in deep shadow, high-contrast chiaroscuro? PASS if the face is clearly divided into a lit half and a shadowed half; FAIL if the lighting is flat or evenly spread. | Specific cinematic effect achieved | The prompt specified camera/lighting: "{prompt_excerpt}". Does the perspective, depth of field, and lighting pattern match what was described? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_ColorPrecision_Easy | Professional Studio | Color Precision | Easy | A bright red sports car parked on a gray asphalt road — the red is vivid and unambiguous | A woman wearing a royal blue evening dress against a neutral background | A field of pure sunflower yellow flowers under a clear blue sky | Is the sports car a vivid, unambiguous red (not orange, maroon, or pink) parked on gray asphalt? PASS if the car is clearly bright red; FAIL if the colour is any other hue or the car is absent. | Is the woman's evening dress a clear royal blue (a vivid medium-to-deep blue, not navy, teal, purple, or cyan) against a neutral background? PASS if the dress is recognisably royal blue; FAIL otherwise. | Are the flowers pure sunflower yellow under a clear blue sky? PASS if the flowers are a saturated, unambiguous yellow (not orange or gold) and the sky is blue; FAIL otherwise. | Color matches name — no ambiguity | The prompt specified exact colors: "{prompt_excerpt}". Do the colors in the image match the described/named colors or hex values? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_ColorPrecision_Medium | Professional Studio | Color Precision | Medium | A bedroom wall painted in #FF6B35 (burnt orange) with white furniture and natural daylight | A forest scene where the foliage is precisely #2D5F2D (dark muted green) with brown tree trunks | A room with walls in #4A90D9 (medium blue) with warm wooden floor and white trim | Is the bedroom wall a burnt-orange color (#FF6B35 — a warm mid orange), paired with white furniture in natural daylight? PASS if the wall is clearly burnt orange (not red, brown, or yellow); FAIL otherwise. | Is the forest foliage a dark, muted green (#2D5F2D — deep desaturated green, not bright or lime) with brown tree trunks? PASS if the leaves read as dark muted green; FAIL if the greens are vivid/light or another hue. | Are the room's walls a medium blue (#4A90D9 — a clear sky-like blue, not navy or teal), with a warm wooden floor and white trim? PASS if the walls are recognisably medium blue; FAIL otherwise. | Measurable closeness to hex value in output | The prompt specified exact colors: "{prompt_excerpt}". Do the colors in the image match the described/named colors or hex values? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_ColorPrecision_Hard | Professional Studio | Color Precision | Hard | A minimalist room with walls in #F5E6CC (warm cream), a sofa in #2C3E50 (dark navy), and accent pillows in #E74C3C (bright red) | A product flat-lay on a #1A1A2E (deep dark blue) background, with a #F0C27F (soft gold) watch and #FFFFFF (pure white) earbuds | A graphic design composition with background #FAFAFA (near-white), main shape #FF4500 (orange-red), and accent lines #333333 (dark charcoal) | Does this minimalist room have warm cream walls (#F5E6CC), a dark navy sofa (#2C3E50), and bright red accent pillows (#E74C3C)? PASS only if all three colors are present and correctly assigned to wall, sofa, and pillows; FAIL if any is missing, wrong, or mis-assigned. | Is this a flat-lay on a deep dark-blue background (#1A1A2E) with a soft-gold watch (#F0C27F) and pure-white earbuds (#FFFFFF)? PASS only if all three colors are present and correctly assigned to background, watch, and earbuds; FAIL otherwise. | Does this graphic composition have a near-white, grey background (#FAFAFA), a main shape in orange-red (#FF4500), and accent lines in dark charcoal (#333333)? PASS only if all three colors are present and correctly assigned to background, main shape, and lines; FAIL otherwise. | All three colors present + correctly assigned | The prompt specified exact colors: "{prompt_excerpt}". Do the colors in the image match the described/named colors or hex values? Rate: PASS or FAIL. | gemma4-26b |
ProfessionalStudio_ColorPrecision_Extreme | Professional Studio | Color Precision | Extreme | A smooth gradient on a wall going from #FF0000 (red) on the left to #0000FF (blue) on the right, transitioning evenly through purple | A sunset sky gradient on a flat surface from #FF6B35 (orange) at the bottom to #1B1464 (deep indigo) at the top | A horizontal banner with a gradient from #00C9FF (cyan) on the left to #92FE9D (mint green) on the right | Does the wall show a smooth horizontal gradient going from red on the left to blue on the right, passing evenly through purple in the middle? PASS if both endpoints (red left, blue right) and the purple mid-transition are correct; FAIL if the direction is wrong, the endpoints differ, or the blend is not smooth. | Does this flat surface show a smooth vertical gradient from warm orange (#FF6B35) at the bottom to deep indigo (#1B1464) at the top, like a sunset sky? PASS if the orange-bottom to indigo-top direction and endpoint colors are correct and evenly blended; FAIL otherwise. | Does this horizontal banner show a smooth gradient from cyan (#00C9FF) on the left to mint green (#92FE9D) on the right? PASS if both endpoints (cyan left, mint-green right) are correct and the blend is smooth and horizontal; FAIL otherwise. | Gradient direction + endpoint colors correct | The prompt specified exact colors: "{prompt_excerpt}". Do the colors in the image match the described/named colors or hex values? Rate: PASS or FAIL. | gemma4-26b |
GraphicalDesign_StyleDiversity_Easy | Graphical design | Style Diversity | Easy | A golden retriever sitting in a garden, painted in the style of an oil painting with visible brushstrokes and rich texture | A golden retriever sitting in a garden, rendered in Japanese anime style with flat colors and large expressive eyes | A golden retriever sitting in a garden, rendered in pixel art style with a 32x32 grid and limited color palette | Does this image show a golden retriever sitting in a garden rendered as an oil painting with clearly visible brushstrokes and rich painterly texture? PASS only if the subject is a golden retriever in a garden AND the oil-painting style (visible brushstrokes, thick textured paint) is clearly present; FAIL otherwise. | Does this image show a golden retriever sitting in a garden drawn in Japanese anime style with flat cel-shaded colors and large expressive eyes? PASS only if the subject is a golden retriever in a garden AND the anime style (flat colors, large expressive eyes) is clearly present; FAIL otherwise. | Does this image show a golden retriever sitting in a garden rendered as pixel art built from visible square pixels on a low-resolution grid with a limited color palette? PASS only if the subject is a golden retriever in a garden AND the pixel-art style (blocky pixels, limited palette) is clearly present; FAIL otherwise... | Style recognizable by any viewer | The prompt asked for a specific art style: "{prompt_excerpt}". Is the requested style clearly recognizable? Are its key visual characteristics present? Rate: PASS or FAIL. | qwen35-122b |
GraphicalDesign_StyleDiversity_Medium | Graphical design | Style Diversity | Medium | A city street scene in the style of Art Deco — geometric shapes, gold and black palette, ornamental symmetry | A woman with a parasol in the style of Impressionism — soft brushstrokes, dappled light, pastel palette | A portrait of a man in Pop Art style — bold outlines, Ben-Day dots, flat vivid primary colors | Is this a city street scene rendered in Art Deco style, showing its signature traits: geometric shapes, a gold-and-black color palette, and ornamental symmetry? PASS only if it is a city street scene AND the Art Deco characteristics are clearly present; FAIL otherwise. | Is this an image of a woman holding a parasol rendered in Impressionist style, showing soft visible brushstrokes, dappled light, and a pastel palette? PASS only if a woman with a parasol is depicted AND the Impressionist characteristics are clearly present; FAIL otherwise. | Is this a portrait of a man rendered in Pop Art style, showing bold black outlines, Ben-Day dots, and flat vivid primary colors? PASS only if it is a portrait of a man AND the Pop Art characteristics are clearly present; FAIL otherwise. | Key visual characteristics of movement present | The prompt asked for a specific art style: "{prompt_excerpt}". Is the requested style clearly recognizable? Are its key visual characteristics present? Rate: PASS or FAIL. | qwen35-122b |
GraphicalDesign_StyleDiversity_Hard | Graphical design | Style Diversity | Hard | A forest rendered in photorealistic watercolor — the detail and lighting of a photo combined with visible watercolor bleeds and paper texture | A woman in a futuristic city rendered in cyberpunk Art Nouveau — neon colors and tech elements combined with flowing organic lines and floral borders | A medieval castle depicted in Japanese ukiyo-e woodblock print style combined with photorealistic sky and lighting | Does this image depict a forest that combines detail and lighting WITH visible watercolor bleeds and paper texture? PASS only if the subject is a forest AND both qualities are simultaneously identifiable (good realism plus watercolor bleeds/paper texture); FAIL if either is missing. | Does this image depict a woman in a futuristic city that fuses cyberpunk elements (neon colors, tech) WITH Art Nouveau elements (flowing organic lines, floral borders)? PASS only if a woman in a futuristic city is shown AND both the cyberpunk and Art Nouveau characteristics are simultaneously identifiable; FAIL if eith... | Does this image depict a medieval castle in Japanese ukiyo-e woodblock-print style combined with a photorealistic sky and lighting? PASS only if a medieval castle is shown AND both the ukiyo-e woodblock style and a photorealistic sky/lighting are simultaneously identifiable; FAIL if either is missing. | Both styles visually identifiable in result | The prompt asked for a specific art style: "{prompt_excerpt}". Is the requested style clearly recognizable? Are its key visual characteristics present? Rate: PASS or FAIL. | qwen35-122b |
GraphicalDesign_StyleDiversity_Extreme | Graphical design | Style Diversity | Extreme | An interior scene rendered in soft brutalism — raw concrete textures combined with rounded pastel furniture and warm diffused light | A still life painted in anxious minimalism — sparse composition with uncomfortable negative space, muted colors, and subtle visual tension | A cityscape rendered in nostalgic futurism — retro-futuristic 1960s space-age aesthetics with a melancholic warm-toned patina | Does this interior scene render 'soft brutalism' — raw exposed concrete textures combined with rounded pastel-colored furniture and warm diffused light? PASS only if it is an interior scene AND it clearly combines raw concrete with rounded pastel furniture and soft warm lighting; FAIL otherwise. | Is this a still life in an 'anxious minimalism' style — a sparse composition with large uncomfortable negative space, a muted color palette, and subtle visual tension? PASS only if it is a minimalist still life with sparse composition, muted colors, and noticeably uneasy negative space; FAIL otherwise. | Is this a cityscape in a 'nostalgic futurism' style — retro-futuristic 1960s space-age aesthetics rendered with a melancholic, warm-toned patina? PASS only if it is a cityscape showing retro 1960s space-age design AND a warm, faded/nostalgic tone; FAIL otherwise. | Result evokes the described feeling/contradiction | The prompt asked for a specific art style: "{prompt_excerpt}". Is the requested style clearly recognizable? Are its key visual characteristics present? Rate: PASS or FAIL. | qwen35-122b |
GraphicalDesign_LayoutDesign_Easy | Graphical design | Layout & Design | Easy | A movie night poster with a large popcorn image in the center, the title 'MOVIE NIGHT' in bold at the top, and the tagline 'Every Friday at 8 PM' at the bottom | An event poster featuring a guitar illustration, the headline 'LIVE MUSIC FEST' in large text, and 'Tickets available now' as a tagline below | A book club poster with an open book illustration, the title 'JOIN THE BOOK CLUB' prominently displayed, and 'Meets every Wednesday' as the tagline | Does this movie-night poster contain ALL of: a large popcorn image in the center, the title 'MOVIE NIGHT' in bold at the top, and the tagline 'Every Friday at 8 PM' at the bottom? PASS only if all three elements are present, correctly placed, and the text is legible and correctly spelled; FAIL if any is missing, mispla... | Does this event poster contain ALL of: a guitar illustration, the headline 'LIVE MUSIC FEST' in large text, and the tagline 'Tickets available now' below it? PASS only if all three elements are present, correctly placed, and the text is legible and correctly spelled; FAIL if any is missing, misplaced, or misspelled. | Does this book-club poster contain ALL of: an open-book illustration, the title 'JOIN THE BOOK CLUB' prominently displayed, and the tagline 'Meets every Wednesday'? PASS only if all three elements are present, correctly placed, and the text is legible and correctly spelled; FAIL if any is missing, misplaced, or misspel... | Visual hierarchy clear + text readable | The prompt described a specific layout: "{prompt_excerpt}". Are all described layout zones/elements present with correct visual hierarchy? Is any text readable and correctly placed? Rate: PASS or FAIL. | qwen3-vl |
GraphicalDesign_LayoutDesign_Medium | Graphical design | Layout & Design | Medium | A landing page design with a hero header image at the top, a headline and subtitle in the middle, a 'Sign Up' call-to-action button, and a footer with three social media icons | A newsletter layout with a banner image header, a two-column body section with text on the left and an image on the right, a highlighted quote block, and a footer with unsubscribe link | A restaurant menu layout with a logo header, three category sections (appetizers, mains, desserts) each with items and prices, and a footer with address and hours | Is the text readable? Does the layout contain ALL of: a hero header image at the top, a headline with a subtitle in the middle, a 'Sign Up' call-to-action button, and a footer with three social-media icons? PASS only if all four zones are present and correctly placed and text readable; FAIL if any is missing. | Is the text readable? Does the layout contain ALL of: a banner image header, a two-column body (text on the left, image on the right), a highlighted quote block, and a footer with an unsubscribe link? PASS only if all are present and placed correctly; FAIL if any is missing. | Does the text is readable? Does the layout contain ALL of: a logo header, three labelled category sections (appetizers, mains, desserts) each with items and prices, and a footer with address and hours? PASS only if all are present; FAIL if any section is missing. | Zones visually separated + content in right zones | The prompt described a specific layout: "{prompt_excerpt}". Are all described layout zones/elements present with correct visual hierarchy? Is any text readable and correctly placed? Rate: PASS or FAIL. | qwen3-vl |
GraphicalDesign_LayoutDesign_Hard | Graphical design | Layout & Design | Hard | A magazine double-page spread with a left page containing a large hero photo, and a right page with a two-column article, a pull quote, a sidebar with author bio, and a page number | A mobile app UI mockup showing a dashboard with a top navigation bar, a search field, a 2x2 grid of feature cards with icons, a list of recent activity items, and a bottom tab bar | A fashion magazine spread with a full-bleed left-page photo, right page with three-column layout, headline, body copy, a smaller inset image, and styled page numbers | Is this a magazine double-page spread where the LEFT page contains a large hero photo, and the RIGHT page contains ALL of: a two-column article, a pull quote, a sidebar with an author bio, and a page number? PASS only if the left hero photo and all four right-page elements are present and correctly placed; FAIL if any ... | Is this a mobile app dashboard UI mockup containing ALL of: a top navigation bar, a search field, a 2x2 grid of feature cards with icons, a list of recent-activity items, and a bottom tab bar? PASS only if all five elements are present and correctly placed; FAIL if any is missing. | Is this a fashion-magazine spread where the LEFT page is a full-bleed photo, and the RIGHT page contains ALL of: a three-column layout, a headline, body copy, a smaller inset image, and styled page numbers? PASS only if the full-bleed left photo and all five right-page elements are present and correctly placed; FAIL if... | Professional-looking layout + all elements present | The prompt described a specific layout: "{prompt_excerpt}". Are all described layout zones/elements present with correct visual hierarchy? Is any text readable and correctly placed? Rate: PASS or FAIL. | qwen3-vl |
GraphicalDesign_DataVisualization_Easy | Graphical design | Data Visualisation | Easy | A bar chart showing monthly sales for January through June with values 12K, 18K, 15K, 22K, 19K, 25K — labeled axes, color-coded bars, and a title 'Monthly Sales 2025' | A pie chart showing market share: Company A 35%, Company B 25%, Company C 20%, Others 20% — each slice a different color with labels and percentages | A simple line graph showing daily website visitors over one week (Mon-Sun): 1200, 1500, 1100, 1800, 2000, 1700, 1400 — with grid lines, axis labels, and title 'Weekly Traffic' | Is this a bar chart titled 'Monthly Sales 2025' showing six labeled bars for January through June with values 12K, 18K, 15K, 22K, 19K, and 25K, with labeled axes and color-coded bars? PASS only if it is a bar chart with all six correctly-labeled months whose bar heights match those values, plus the title and labeled ax... | Is this a pie chart showing market share with four labeled slices: Company A 35%, Company B 25%, Company C 20%, and Others 20%, each a different color with its label and percentage shown? PASS only if it is a pie chart with those four slices, correct percentages, and visible labels; FAIL if the chart type, segments, or... | Is this a line graph titled 'Weekly Traffic' showing daily website visitors for Monday through Sunday with values 1200, 1500, 1100, 1800, 2000, 1700, and 1400, including grid lines and axis labels? PASS only if it is a line graph with all seven labeled days whose points match those values, plus the title and axis label... | Numbers correct + layout logical + readable | The prompt asked for a data visualization: "{prompt_excerpt}". Are the numbers, labels, and visual elements correct and readable? Does the chart type match what was requested? Rate: PASS or FAIL. | qwen3-vl |
ImageBench — 50 Text-to-Image Models Judged by VLMs on 192 Prompts
Reproducibility dataset for imagebench.ai: the 192-prompt V1.2 benchmark, per-(model, prompt) VLM verdicts, and per-model aggregate scores for 50 text-to-image models.
Live leaderboard + every generated image: https://imagebench.ai Methodology: https://imagebench.ai/methodology-v1 Reproducibility repo: https://github.com/dh7/image-bench-ai
What's in this dataset
| File | Rows | Description |
|---|---|---|
prompts.csv |
64 tests × 3 variants | The V1.2 benchmark: prompt variants, category/subcategory/difficulty, VLM judge questions, evaluation criteria, VLM-routing assignment. |
verdicts.csv |
9,600 (50 models × 192 prompts) | Per-image pass/fail verdict, VLM judge reasoning, HPS aesthetic score, VLM model used. |
scores.csv |
50 | Per-model aggregate: Overall (0-100), pass rate, aesthetic Estimated Preference Score, per-category pass rates. |
Column reference
verdicts.csv
model_slug— canonical model identifier (matches URLs on imagebench.ai)model_name— human-readable namecategory/subcategory/difficulty— benchmark taxonomyprompt_id— matches image filename on imagebench.ai/galleryprompt— the actual prompt sent to the modelverdict—PASSorFAILfrom the VLM judgevlm— which VLM produced the verdict (Qwen 3.5 122B / Gemini 3.1 Pro / etc.)judge_response— the VLM's textual reasoning for the verdicthps_mu— HPSv3 aesthetic score (higher = judged more aesthetically pleasing by human-preference model)hps_pref— pairwise HPS preference metadata
scores.csv
model_slug,model_nameoverall— 0-100 blended score (capability + aesthetic preference)pass_rate— % of prompts passing the VLM verdicteps— Estimated Preference Score (aesthetic, based on HPSv3)pass_<category>— pass rate per category (text rendering, spatial reasoning, human realism, truthfulness, studio quality, graphical design)
How this was produced
- Prompts: 64 tests × 3 variants, hand-curated across 6 capability categories with binary judge questions per test.
- Generation: each model produces 192 images (one per prompt) via its native API — no cherry-picking, no re-runs.
- Judging: VLM-routed — Qwen 3.5 122B handles most categories, with per-category routing to specialists (Gemini 3.1 Pro for hands, etc.) based on calibration.
- Aesthetic score: HPSv3 (Human Preference Score v3) applied to every generated image.
- Overall: 0.5 × normalized pass rate + 0.5 × normalized EPS.
Full methodology + calibration study: https://imagebench.ai/methodology-v1
Known limitations
- Text-to-image only — no editing, inpainting, video, long-form coherence.
- Prompt bias — the 192-prompt set is weighted toward tests that separate models. "Every model passes" mundane prompts are under-represented, so pass rates are lower than typical real-world use.
- VLM judge bias — VLMs have their own systematic errors. We validate on a hands-specific calibration blog: https://imagebench.ai/blog/hands-benchmark-qwen35-122b
- Category weighting — Overall gives each category equal weight, but "text rendering" is inherently harder to pass than "professional studio quality". Interpret cross-category comparisons carefully.
Citation
If you use this dataset in research or writing, please cite:
ImageBench: 50 Text-to-Image Models Judged by VLMs on 192 Prompts
https://imagebench.ai
License
CC-BY 4.0 — free to use with attribution.
Feedback
Issues, corrections, or methodology critique: open an issue at https://github.com/dh7/image-bench-ai or leave a comment on this dataset.
- Downloads last month
- -