Merge branch 'main' into space-txt2img

README.md

@@ -127,6 +127,13 @@ docker build -t lcm-live .
 docker run -ti -p 7860:7860 --gpus all lcm-live
 ```
 
+reuse models data from host to avoid downloading them again, you can change `~/.cache/huggingface` to any other directory, but if you use hugingface-cli locally, you can share the same cache
+
+```bash
+docker run -ti -p 7860:7860 -e HF_HOME=/data -v ~/.cache/huggingface:/data  --gpus all lcm-live
+```
+ 
+
 or with environment variables
 
 ```bash

app_init.py

@@ -3,6 +3,7 @@ from fastapi.responses import StreamingResponse, JSONResponse
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.staticfiles import StaticFiles
 from fastapi import Request
+import markdown2
 
 import logging
 import traceback
@@ -13,6 +14,7 @@ import time
 from types import SimpleNamespace
 from util import pil_to_frame, bytes_to_pil, is_firefox
 import asyncio
+import os
 
 
 def init_app(app: FastAPI, user_data: UserData, args: Args, pipeline):
@@ -41,11 +43,7 @@ def init_app(app: FastAPI, user_data: UserData, args: Args, pipeline):
             await websocket.send_json(
                 {"status": "connected", "message": "Connected", "userId": str(user_id)}
             )
-            await websocket.send_json(
-                {
-                    "status": "send_frame",
-                }
-            )
+            await websocket.send_json({"status": "send_frame"})
             await handle_websocket_data(user_id, websocket)
         except WebSocketDisconnect as e:
             logging.error(f"WebSocket Error: {e}, {user_id}")
@@ -71,13 +69,12 @@ def init_app(app: FastAPI, user_data: UserData, args: Args, pipeline):
                 params = SimpleNamespace(**params.dict())
                 if info.input_mode == "image":
                     image_data = await websocket.receive_bytes()
+                    if len(image_data) == 0:
+                        await websocket.send_json({"status": "send_frame"})
+                        continue
                     params.image = bytes_to_pil(image_data)
                 await user_data.update_data(user_id, params)
-                await websocket.send_json(
-                    {
-                        "status": "wait",
-                    }
-                )
+                await websocket.send_json({"status": "wait"})
                 if args.timeout > 0 and time.time() - last_time > args.timeout:
                     await websocket.send_json(
                         {
@@ -110,11 +107,7 @@ def init_app(app: FastAPI, user_data: UserData, args: Args, pipeline):
                 while True:
                     params = await user_data.get_latest_data(user_id)
                     if not vars(params) or params.__dict__ == last_params.__dict__:
-                        await websocket.send_json(
-                            {
-                                "status": "send_frame",
-                            }
-                        )
+                        await websocket.send_json({"status": "send_frame"})
                         await asyncio.sleep(0.1)
                         continue
 
@@ -143,14 +136,22 @@ def init_app(app: FastAPI, user_data: UserData, args: Args, pipeline):
     # route to setup frontend
     @app.get("/settings")
     async def settings():
-        info = pipeline.Info.schema()
+        info_schema = pipeline.Info.schema()
+        info = pipeline.Info()
+        if info.page_content:
+            page_content = markdown2.markdown(info.page_content)
+
         input_params = pipeline.InputParams.schema()
         return JSONResponse(
             {
-                "info": info,
+                "info": info_schema,
                 "input_params": input_params,
                 "max_queue_size": args.max_queue_size,
+                "page_content": page_content if info.page_content else "",
             }
         )
 
+    if not os.path.exists("public"):
+        os.makedirs("public")
+
     app.mount("/", StaticFiles(directory="public", html=True), name="public")

frontend/src/lib/components/MediaListSwitcher.svelte

@@ -18,17 +18,19 @@
 <div class="flex items-center justify-center text-xs">
   <button
     title="Share your screen"
-    class="border-1 my-1 block cursor-pointer rounded-md border-gray-500 border-opacity-50 bg-slate-100 bg-opacity-30 p-[2px] font-medium text-white"
+    class="border-1 my-1 flex cursor-pointer gap-1 rounded-md border-gray-500 border-opacity-50 bg-slate-100 bg-opacity-30 p-1 font-medium text-white"
     on:click={() => mediaStreamActions.startScreenCapture()}
   >
-    <Screen classList={'w-100'} />
+    <span>Share</span>
+
+    <Screen classList={''} />
   </button>
   {#if $mediaDevices}
     <select
       bind:value={deviceId}
       on:change={() => mediaStreamActions.switchCamera(deviceId)}
       id="devices-list"
-      class="border-1 block cursor-pointer rounded-md border-gray-800 border-opacity-50 bg-slate-100 bg-opacity-30 p-[2px] font-medium text-white"
+      class="border-1 block cursor-pointer rounded-md border-gray-800 border-opacity-50 bg-slate-100 bg-opacity-30 p-1 font-medium text-white"
     >
       {#each $mediaDevices as device, i}
         <option value={device.deviceId}>{device.label}</option>

frontend/src/lib/components/PipelineOptions.svelte

@@ -1,6 +1,5 @@
 <script lang="ts">
-  import { createEventDispatcher } from 'svelte';
-  import type { FieldProps } from '$lib/types';
+  import type { Fields } from '$lib/types';
   import { FieldType } from '$lib/types';
   import InputRange from './InputRange.svelte';
   import SeedInput from './SeedInput.svelte';
@@ -9,10 +8,10 @@
   import Selectlist from './Selectlist.svelte';
   import { pipelineValues } from '$lib/store';
 
-  export let pipelineParams: FieldProps[];
+  export let pipelineParams: Fields;
 
-  $: advanceOptions = pipelineParams?.filter((e) => e?.hide == true);
-  $: featuredOptions = pipelineParams?.filter((e) => e?.hide !== true);
+  $: advanceOptions = Object.values(pipelineParams)?.filter((e) => e?.hide == true);
+  $: featuredOptions = Object.values(pipelineParams)?.filter((e) => e?.hide !== true);
 </script>
 
 <div class="flex flex-col gap-3">
@@ -37,7 +36,7 @@
   <details>
     <summary class="cursor-pointer font-medium">Advanced Options</summary>
     <div
-      class="grid grid-cols-1 items-center gap-3 {pipelineParams.length > 5
+      class="grid grid-cols-1 items-center gap-3 {Object.values(pipelineParams).length > 5
         ? 'sm:grid-cols-2'
         : ''}"
     >

frontend/src/lib/components/VideoInput.svelte

@@ -10,21 +10,24 @@
     mediaDevices
   } from '$lib/mediaStream';
   import MediaListSwitcher from './MediaListSwitcher.svelte';
+  export let width = 512;
+  export let height = 512;
+  const size = { width, height };
 
   let videoEl: HTMLVideoElement;
   let canvasEl: HTMLCanvasElement;
   let ctx: CanvasRenderingContext2D;
   let videoFrameCallbackId: number;
-  const WIDTH = 768;
-  const HEIGHT = 768;
+
   // ajust the throttle time to your needs
   const THROTTLE_TIME = 1000 / 15;
   let selectedDevice: string = '';
+  let videoIsReady = false;
 
   onMount(() => {
     ctx = canvasEl.getContext('2d') as CanvasRenderingContext2D;
-    canvasEl.width = WIDTH;
-    canvasEl.height = HEIGHT;
+    canvasEl.width = size.width;
+    canvasEl.height = size.height;
   });
   $: {
     console.log(selectedDevice);
@@ -44,35 +47,34 @@
     }
     const videoWidth = videoEl.videoWidth;
     const videoHeight = videoEl.videoHeight;
-    const blob = await grapCropBlobImg(
-      videoEl,
-      videoWidth / 2 - WIDTH / 2,
-      videoHeight / 2 - HEIGHT / 2,
-      WIDTH,
-      HEIGHT
-    );
-
+    let height0 = videoHeight;
+    let width0 = videoWidth;
+    let x0 = 0;
+    let y0 = 0;
+    if (videoWidth > videoHeight) {
+      width0 = videoHeight;
+      x0 = (videoWidth - videoHeight) / 2;
+    } else {
+      height0 = videoWidth;
+      y0 = (videoHeight - videoWidth) / 2;
+    }
+    ctx.drawImage(videoEl, x0, y0, width0, height0, 0, 0, size.width, size.height);
+    const blob = await new Promise<Blob>((resolve) => {
+      canvasEl.toBlob(
+        (blob) => {
+          resolve(blob as Blob);
+        },
+        'image/jpeg',
+        1
+      );
+    });
     onFrameChangeStore.set({ blob });
     videoFrameCallbackId = videoEl.requestVideoFrameCallback(onFrameChange);
   }
 
-  $: if ($mediaStreamStatus == MediaStreamStatusEnum.CONNECTED) {
+  $: if ($mediaStreamStatus == MediaStreamStatusEnum.CONNECTED && videoIsReady) {
     videoFrameCallbackId = videoEl.requestVideoFrameCallback(onFrameChange);
   }
-  async function grapCropBlobImg(
-    video: HTMLVideoElement,
-    x: number,
-    y: number,
-    width: number,
-    height: number
-  ) {
-    const canvas = new OffscreenCanvas(width, height);
-
-    const ctx = canvas.getContext('2d') as OffscreenCanvasRenderingContext2D;
-    ctx.drawImage(video, x, y, width, height, 0, 0, width, height);
-    const blob = await canvas.convertToBlob({ type: 'image/jpeg', quality: 1 });
-    return blob;
-  }
 </script>
 
 <div class="relative mx-auto max-w-lg overflow-hidden rounded-lg border border-slate-300">
@@ -85,6 +87,9 @@
     <video
       class="pointer-events-none aspect-square w-full object-cover"
       bind:this={videoEl}
+      on:loadeddata={() => {
+        videoIsReady = true;
+      }}
       playsinline
       autoplay
       muted

frontend/src/lib/types.ts

@@ -11,6 +11,11 @@ export const enum PipelineMode {
     TEXT = "text",
 }
 
+
+export interface Fields {
+    [key: string]: FieldProps;
+}
+
 export interface FieldProps {
     default: number | string;
     max?: number;

frontend/src/routes/+page.svelte

@@ -1,6 +1,6 @@
 <script lang="ts">
   import { onMount } from 'svelte';
-  import type { FieldProps, PipelineInfo } from '$lib/types';
+  import type { Fields, PipelineInfo } from '$lib/types';
   import { PipelineMode } from '$lib/types';
   import ImagePlayer from '$lib/components/ImagePlayer.svelte';
   import VideoInput from '$lib/components/VideoInput.svelte';
@@ -11,8 +11,9 @@
   import { mediaStreamActions, onFrameChangeStore } from '$lib/mediaStream';
   import { getPipelineValues, deboucedPipelineValues } from '$lib/store';
 
-  let pipelineParams: FieldProps[];
+  let pipelineParams: Fields;
   let pipelineInfo: PipelineInfo;
+  let pageContent: string;
   let isImageMode: boolean = false;
   let maxQueueSize: number = 0;
   let currentQueueSize: number = 0;
@@ -22,11 +23,12 @@
 
   async function getSettings() {
     const settings = await fetch('/settings').then((r) => r.json());
-    pipelineParams = Object.values(settings.input_params.properties);
+    pipelineParams = settings.input_params.properties;
     pipelineInfo = settings.info.properties;
     isImageMode = pipelineInfo.input_mode.default === PipelineMode.IMAGE;
     maxQueueSize = settings.max_queue_size;
-    pipelineParams = pipelineParams.filter((e) => e?.disabled !== true);
+    pageContent = settings.page_content;
+    console.log(pipelineParams);
     if (maxQueueSize > 0) {
       getQueueSize();
       setInterval(() => {
@@ -68,33 +70,17 @@
   }
 </script>
 
+<svelte:head>
+  <script
+    src="https://cdnjs.cloudflare.com/ajax/libs/iframe-resizer/4.3.9/iframeResizer.contentWindow.min.js"
+  ></script>
+</svelte:head>
+
 <main class="container mx-auto flex max-w-5xl flex-col gap-3 px-4 py-4">
   <article class="text-center">
-    <h1 class="text-3xl font-bold">Real-Time Latent Consistency Model</h1>
-    {#if pipelineInfo?.title?.default}
-      <h3 class="text-xl font-bold">{pipelineInfo?.title?.default}</h3>
+    {#if pageContent}
+      {@html pageContent}
     {/if}
-    <p class="text-sm">
-      This demo showcases
-      <a
-        href="https://huggingface.co/blog/lcm_lora"
-        target="_blank"
-        class="text-blue-500 underline hover:no-underline">LCM LoRA</a
-      >
-      Image to Image pipeline using
-      <a
-        href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
-        target="_blank"
-        class="text-blue-500 underline hover:no-underline">Diffusers</a
-      > with a MJPEG stream server.
-    </p>
-    <p class="text-sm text-gray-500">
-      Change the prompt to generate different images, accepts <a
-        href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
-        target="_blank"
-        class="text-blue-500 underline hover:no-underline">Compel</a
-      > syntax.
-    </p>
     {#if maxQueueSize > 0}
       <p class="text-sm">
         There are <span id="queue_size" class="font-bold">{currentQueueSize}</span>
@@ -111,7 +97,10 @@
     <article class="my-3 grid grid-cols-1 gap-3 sm:grid-cols-2">
       {#if isImageMode}
         <div class="sm:col-start-1">
-          <VideoInput></VideoInput>
+          <VideoInput
+            width={Number(pipelineParams.width.default)}
+            height={Number(pipelineParams.height.default)}
+          ></VideoInput>
         </div>
       {/if}
       <div class={isImageMode ? 'sm:col-start-2' : 'col-span-2'}>

frontend/tailwind.config.js

@@ -1,6 +1,6 @@
 /** @type {import('tailwindcss').Config} */
 export default {
-  content: ['./src/**/*.{html,js,svelte,ts}'],
+  content: ['./src/**/*.{html,js,svelte,ts}', '../pipelines/**/*.py'],
   theme: {
     extend: {}
   },

pipelines/controlnelSD21Turbo.py

@@ -0,0 +1,260 @@
+from diffusers import (
+    StableDiffusionControlNetImg2ImgPipeline,
+    ControlNetModel,
+    LCMScheduler,
+    AutoencoderTiny,
+)
+from compel import Compel
+import torch
+from pipelines.utils.canny_gpu import SobelOperator
+
+try:
+    import intel_extension_for_pytorch as ipex  # type: ignore
+except:
+    pass
+
+import psutil
+from config import Args
+from pydantic import BaseModel, Field
+from PIL import Image
+import math
+import time
+
+#
+taesd_model = "madebyollin/taesd"
+controlnet_model = "thibaud/controlnet-sd21-canny-diffusers"
+base_model = "stabilityai/sd-turbo"
+
+default_prompt = "Portrait of The Joker halloween costume, face painting, with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time SDv2.1 Turbo</h1>
+<h3 class="text-xl font-bold">Image-to-Image ControlNet</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/stabilityai/sd-turbo"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">SD Turbo</a>
+Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/sdxl_turbo"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
+
+
+class Pipeline:
+    class Info(BaseModel):
+        name: str = "controlnet+sd15Turbo"
+        title: str = "SDv1.5 Turbo + Controlnet"
+        description: str = "Generates an image from a text prompt"
+        input_mode: str = "image"
+        page_content: str = page_content
+
+    class InputParams(BaseModel):
+        prompt: str = Field(
+            default_prompt,
+            title="Prompt",
+            field="textarea",
+            id="prompt",
+        )
+        seed: int = Field(
+            4402026899276587, min=0, title="Seed", field="seed", hide=True, id="seed"
+        )
+        steps: int = Field(
+            1, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
+        )
+        width: int = Field(
+            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+        )
+        height: int = Field(
+            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+        )
+        guidance_scale: float = Field(
+            1.21,
+            min=0,
+            max=10,
+            step=0.001,
+            title="Guidance Scale",
+            field="range",
+            hide=True,
+            id="guidance_scale",
+        )
+        strength: float = Field(
+            0.8,
+            min=0.10,
+            max=1.0,
+            step=0.001,
+            title="Strength",
+            field="range",
+            hide=True,
+            id="strength",
+        )
+        controlnet_scale: float = Field(
+            0.2,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Controlnet Scale",
+            field="range",
+            hide=True,
+            id="controlnet_scale",
+        )
+        controlnet_start: float = Field(
+            0.0,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Controlnet Start",
+            field="range",
+            hide=True,
+            id="controlnet_start",
+        )
+        controlnet_end: float = Field(
+            1.0,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Controlnet End",
+            field="range",
+            hide=True,
+            id="controlnet_end",
+        )
+        canny_low_threshold: float = Field(
+            0.31,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Canny Low Threshold",
+            field="range",
+            hide=True,
+            id="canny_low_threshold",
+        )
+        canny_high_threshold: float = Field(
+            0.125,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Canny High Threshold",
+            field="range",
+            hide=True,
+            id="canny_high_threshold",
+        )
+        debug_canny: bool = Field(
+            False,
+            title="Debug Canny",
+            field="checkbox",
+            hide=True,
+            id="debug_canny",
+        )
+
+    def __init__(self, args: Args, device: torch.device, torch_dtype: torch.dtype):
+        controlnet_canny = ControlNetModel.from_pretrained(
+            controlnet_model, torch_dtype=torch_dtype
+        ).to(device)
+
+        self.pipes = {}
+
+        if args.safety_checker:
+            self.pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
+                base_model,
+                controlnet=controlnet_canny,
+            )
+        else:
+            self.pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
+                base_model,
+                controlnet=controlnet_canny,
+                safety_checker=None,
+            )
+
+        if args.use_taesd:
+            self.pipe.vae = AutoencoderTiny.from_pretrained(
+                taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+            ).to(device)
+        self.canny_torch = SobelOperator(device=device)
+
+        self.pipe.scheduler = LCMScheduler.from_config(self.pipe.scheduler.config)
+        self.pipe.set_progress_bar_config(disable=True)
+        self.pipe.to(device=device, dtype=torch_dtype).to(device)
+        if device.type != "mps":
+            self.pipe.unet.to(memory_format=torch.channels_last)
+
+        if psutil.virtual_memory().total < 64 * 1024**3:
+            self.pipe.enable_attention_slicing()
+
+        self.pipe.compel_proc = Compel(
+            tokenizer=self.pipe.tokenizer,
+            text_encoder=self.pipe.text_encoder,
+            truncate_long_prompts=True,
+        )
+        if args.use_taesd:
+            self.pipe.vae = AutoencoderTiny.from_pretrained(
+                taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+            ).to(device)
+
+        if args.torch_compile:
+            self.pipe.unet = torch.compile(
+                self.pipe.unet, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe.vae = torch.compile(
+                self.pipe.vae, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe(
+                prompt="warmup",
+                image=[Image.new("RGB", (768, 768))],
+                control_image=[Image.new("RGB", (768, 768))],
+            )
+
+    def predict(self, params: "Pipeline.InputParams") -> Image.Image:
+        generator = torch.manual_seed(params.seed)
+        prompt_embeds = self.pipe.compel_proc(params.prompt)
+        control_image = self.canny_torch(
+            params.image, params.canny_low_threshold, params.canny_high_threshold
+        )
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
+        last_time = time.time()
+        results = self.pipe(
+            image=params.image,
+            control_image=control_image,
+            prompt_embeds=prompt_embeds,
+            generator=generator,
+            strength=strength,
+            num_inference_steps=steps,
+            guidance_scale=params.guidance_scale,
+            width=params.width,
+            height=params.height,
+            output_type="pil",
+            controlnet_conditioning_scale=params.controlnet_scale,
+            control_guidance_start=params.controlnet_start,
+            control_guidance_end=params.controlnet_end,
+        )
+        print(f"Time taken: {time.time() - last_time}")
+
+        nsfw_content_detected = (
+            results.nsfw_content_detected[0]
+            if "nsfw_content_detected" in results
+            else False
+        )
+        if nsfw_content_detected:
+            return None
+        result_image = results.images[0]
+        if params.debug_canny:
+            # paste control_image on top of result_image
+            w0, h0 = (200, 200)
+            control_image = control_image.resize((w0, h0))
+            w1, h1 = result_image.size
+            result_image.paste(control_image, (w1 - w0, h1 - h0))
+
+        return result_image

pipelines/controlnet.py

@@ -16,12 +16,38 @@ import psutil
 from config import Args
 from pydantic import BaseModel, Field
 from PIL import Image
+import math
 
 base_model = "SimianLuo/LCM_Dreamshaper_v7"
 taesd_model = "madebyollin/taesd"
 controlnet_model = "lllyasviel/control_v11p_sd15_canny"
 
 default_prompt = "Portrait of The Terminator with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model</h1>
+<h3 class="text-xl font-bold">LCM + Controlnet Canny</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/blog/lcm_lora"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM LoRA</a
+    >
+    ControlNet + Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
 
 
 class Pipeline:
@@ -30,6 +56,7 @@ class Pipeline:
         title: str = "LCM + Controlnet"
         description: str = "Generates an image from a text prompt"
         input_mode: str = "image"
+        page_content: str = page_content
 
     class InputParams(BaseModel):
         prompt: str = Field(
@@ -42,13 +69,13 @@ class Pipeline:
             2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
         )
         steps: int = Field(
-            4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
+            4, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
-            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+            768, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
         )
         height: int = Field(
-            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+            768, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
         )
         guidance_scale: float = Field(
             0.2,
@@ -145,7 +172,7 @@ class Pipeline:
         if args.use_taesd:
             self.pipe.vae = AutoencoderTiny.from_pretrained(
                 taesd_model, torch_dtype=torch_dtype, use_safetensors=True
-            )
+            ).to(device)
         self.canny_torch = SobelOperator(device=device)
         self.pipe.set_progress_bar_config(disable=True)
         self.pipe.to(device=device, dtype=torch_dtype)
@@ -182,14 +209,18 @@ class Pipeline:
         control_image = self.canny_torch(
             params.image, params.canny_low_threshold, params.canny_high_threshold
         )
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
 
         results = self.pipe(
             image=params.image,
             control_image=control_image,
             prompt_embeds=prompt_embeds,
             generator=generator,
-            strength=params.strength,
-            num_inference_steps=params.steps,
+            strength=strength,
+            num_inference_steps=steps,
             guidance_scale=params.guidance_scale,
             width=params.width,
             height=params.height,

pipelines/controlnetLoraSD15.py

@@ -2,6 +2,7 @@ from diffusers import (
     StableDiffusionControlNetImg2ImgPipeline,
     ControlNetModel,
     LCMScheduler,
+    AutoencoderTiny,
 )
 from compel import Compel
 import torch
@@ -16,6 +17,7 @@ import psutil
 from config import Args
 from pydantic import BaseModel, Field
 from PIL import Image
+import math
 
 taesd_model = "madebyollin/taesd"
 controlnet_model = "lllyasviel/control_v11p_sd15_canny"
@@ -26,17 +28,40 @@ base_models = {
     "nitrosocke/mo-di-diffusion": "modern disney style",
 }
 lcm_lora_id = "latent-consistency/lcm-lora-sdv1-5"
-
-
 default_prompt = "Portrait of The Terminator with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model SDv1.5</h1>
+<h3 class="text-xl font-bold">LCM + LoRA + Controlnet + Canny</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/blog/lcm_lora"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM LoRA</a>
++ ControlNet + Image to Imasge pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
 
 
 class Pipeline:
     class Info(BaseModel):
         name: str = "controlnet+loras+sd15"
-        title: str = "LCM + LoRA + Controlnet "
+        title: str = "LCM + LoRA + Controlnet"
         description: str = "Generates an image from a text prompt"
         input_mode: str = "image"
+        page_content: str = page_content
 
     class InputParams(BaseModel):
         prompt: str = Field(
@@ -45,24 +70,24 @@ class Pipeline:
             field="textarea",
             id="prompt",
         )
-        model_id: str = Field(
+        base_model_id: str = Field(
             "plasmo/woolitize",
             title="Base Model",
             values=list(base_models.keys()),
             field="select",
-            id="model_id",
+            id="base_model_id",
         )
         seed: int = Field(
             2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
         )
         steps: int = Field(
-            4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
+            4, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
-            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+            768, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
         )
         height: int = Field(
-            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+            768, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
         )
         guidance_scale: float = Field(
             0.2,
@@ -150,20 +175,20 @@ class Pipeline:
         self.pipes = {}
 
         if args.safety_checker:
-            for model_id in base_models.keys():
+            for base_model_id in base_models.keys():
                 pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
-                    model_id,
+                    base_model_id,
                     controlnet=controlnet_canny,
                 )
-            self.pipes[model_id] = pipe
+            self.pipes[base_model_id] = pipe
         else:
-            for model_id in base_models.keys():
+            for base_model_id in base_models.keys():
                 pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
-                    model_id,
+                    base_model_id,
                     safety_checker=None,
                     controlnet=controlnet_canny,
                 )
-                self.pipes[model_id] = pipe
+                self.pipes[base_model_id] = pipe
 
         self.canny_torch = SobelOperator(device=device)
 
@@ -177,6 +202,11 @@ class Pipeline:
             if psutil.virtual_memory().total < 64 * 1024**3:
                 pipe.enable_attention_slicing()
 
+            if args.use_taesd:
+                pipe.vae = AutoencoderTiny.from_pretrained(
+                    taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+                ).to(device)
+
             # Load LCM LoRA
             pipe.load_lora_weights(lcm_lora_id, adapter_name="lcm")
             pipe.compel_proc = Compel(
@@ -199,23 +229,26 @@ class Pipeline:
 
     def predict(self, params: "Pipeline.InputParams") -> Image.Image:
         generator = torch.manual_seed(params.seed)
-        print(f"Using model: {params.model_id}")
-        pipe = self.pipes[params.model_id]
+        pipe = self.pipes[params.base_model_id]
 
-        activation_token = base_models[params.model_id]
+        activation_token = base_models[params.base_model_id]
         prompt = f"{activation_token} {params.prompt}"
         prompt_embeds = pipe.compel_proc(prompt)
         control_image = self.canny_torch(
             params.image, params.canny_low_threshold, params.canny_high_threshold
         )
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
 
         results = pipe(
             image=params.image,
             control_image=control_image,
             prompt_embeds=prompt_embeds,
             generator=generator,
-            strength=params.strength,
-            num_inference_steps=params.steps,
+            strength=strength,
+            num_inference_steps=steps,
             guidance_scale=params.guidance_scale,
             width=params.width,
             height=params.height,

pipelines/controlnetLoraSDXL.py

@@ -3,6 +3,7 @@ from diffusers import (
     ControlNetModel,
     LCMScheduler,
     AutoencoderKL,
+    AutoencoderTiny,
 )
 from compel import Compel, ReturnedEmbeddingsType
 import torch
@@ -17,30 +18,49 @@ import psutil
 from config import Args
 from pydantic import BaseModel, Field
 from PIL import Image
+import math
 
 controlnet_model = "diffusers/controlnet-canny-sdxl-1.0"
 model_id = "stabilityai/stable-diffusion-xl-base-1.0"
 lcm_lora_id = "latent-consistency/lcm-lora-sdxl"
-
-# # base model with activation token, it will prepend the prompt with the activation token
-base_models = {
-    "plasmo/woolitize": "woolitize",
-    "nitrosocke/Ghibli-Diffusion": "ghibli style",
-    "nitrosocke/mo-di-diffusion": "modern disney style",
-}
-# lcm_lora_id = "latent-consistency/lcm-lora-sdv1-5"
+taesd_model = "madebyollin/taesdxl"
 
 
 default_prompt = "Portrait of The Terminator with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
 default_negative_prompt = "blurry, low quality, render, 3D, oversaturated"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model SDXL</h1>
+<h3 class="text-xl font-bold">SDXL + LCM + LoRA + Controlnet</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/blog/lcm_lora"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM LoRA</a>
++ SDXL + Controlnet + Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
 
 
 class Pipeline:
     class Info(BaseModel):
         name: str = "controlnet+loras+sdxl"
-        title: str = "SDXL + LCM + LoRA + Controlnet "
+        title: str = "SDXL + LCM + LoRA + Controlnet"
         description: str = "Generates an image from a text prompt"
         input_mode: str = "image"
+        page_content: str = page_content
 
     class InputParams(BaseModel):
         prompt: str = Field(
@@ -60,13 +80,13 @@ class Pipeline:
             2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
         )
         steps: int = Field(
-            4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
+            2, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
-            768, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+            1024, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
         )
         height: int = Field(
-            768, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+            1024, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
         )
         guidance_scale: float = Field(
             1.0,
@@ -79,10 +99,10 @@ class Pipeline:
             id="guidance_scale",
         )
         strength: float = Field(
-            0.5,
+            1,
             min=0.25,
             max=1.0,
-            step=0.001,
+            step=0.0001,
             title="Strength",
             field="range",
             hide=True,
@@ -191,6 +211,10 @@ class Pipeline:
             returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
             requires_pooled=[False, True],
         )
+        if args.use_taesd:
+            self.pipe.vae = AutoencoderTiny.from_pretrained(
+                taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+            ).to(device)
 
         if args.torch_compile:
             self.pipe.unet = torch.compile(
@@ -214,6 +238,10 @@ class Pipeline:
         control_image = self.canny_torch(
             params.image, params.canny_low_threshold, params.canny_high_threshold
         )
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
 
         results = self.pipe(
             image=params.image,
@@ -223,8 +251,8 @@ class Pipeline:
             negative_prompt_embeds=prompt_embeds[1:2],
             negative_pooled_prompt_embeds=pooled_prompt_embeds[1:2],
             generator=generator,
-            strength=params.strength,
-            num_inference_steps=params.steps,
+            strength=strength,
+            num_inference_steps=steps,
             guidance_scale=params.guidance_scale,
             width=params.width,
             height=params.height,

pipelines/controlnetSDXLTurbo.py

@@ -0,0 +1,268 @@
+from diffusers import (
+    StableDiffusionXLControlNetImg2ImgPipeline,
+    ControlNetModel,
+    AutoencoderKL,
+    AutoencoderTiny,
+)
+from compel import Compel, ReturnedEmbeddingsType
+import torch
+from pipelines.utils.canny_gpu import SobelOperator
+
+try:
+    import intel_extension_for_pytorch as ipex  # type: ignore
+except:
+    pass
+
+import psutil
+from config import Args
+from pydantic import BaseModel, Field
+from PIL import Image
+import math
+
+controlnet_model = "diffusers/controlnet-canny-sdxl-1.0"
+model_id = "stabilityai/sdxl-turbo"
+taesd_model = "madebyollin/taesdxl"
+
+default_prompt = "Portrait of The Terminator with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
+default_negative_prompt = "blurry, low quality, render, 3D, oversaturated"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time SDXL Turbo</h1>
+<h3 class="text-xl font-bold">Image-to-Image ControlNet</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/stabilityai/sdxl-turbo"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">SDXL Turbo</a>
+Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/sdxl_turbo"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
+
+
+class Pipeline:
+    class Info(BaseModel):
+        name: str = "controlnet+SDXL+Turbo"
+        title: str = "SDXL Turbo + Controlnet"
+        description: str = "Generates an image from a text prompt"
+        input_mode: str = "image"
+        page_content: str = page_content
+
+    class InputParams(BaseModel):
+        prompt: str = Field(
+            default_prompt,
+            title="Prompt",
+            field="textarea",
+            id="prompt",
+        )
+        negative_prompt: str = Field(
+            default_negative_prompt,
+            title="Negative Prompt",
+            field="textarea",
+            id="negative_prompt",
+            hide=True,
+        )
+        seed: int = Field(
+            2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
+        )
+        steps: int = Field(
+            2, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
+        )
+        width: int = Field(
+            1024, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+        )
+        height: int = Field(
+            1024, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+        )
+        guidance_scale: float = Field(
+            1.0,
+            min=0,
+            max=10,
+            step=0.001,
+            title="Guidance Scale",
+            field="range",
+            hide=True,
+            id="guidance_scale",
+        )
+        strength: float = Field(
+            0.5,
+            min=0.25,
+            max=1.0,
+            step=0.001,
+            title="Strength",
+            field="range",
+            hide=True,
+            id="strength",
+        )
+        controlnet_scale: float = Field(
+            0.5,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Controlnet Scale",
+            field="range",
+            hide=True,
+            id="controlnet_scale",
+        )
+        controlnet_start: float = Field(
+            0.0,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Controlnet Start",
+            field="range",
+            hide=True,
+            id="controlnet_start",
+        )
+        controlnet_end: float = Field(
+            1.0,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Controlnet End",
+            field="range",
+            hide=True,
+            id="controlnet_end",
+        )
+        canny_low_threshold: float = Field(
+            0.31,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Canny Low Threshold",
+            field="range",
+            hide=True,
+            id="canny_low_threshold",
+        )
+        canny_high_threshold: float = Field(
+            0.125,
+            min=0,
+            max=1.0,
+            step=0.001,
+            title="Canny High Threshold",
+            field="range",
+            hide=True,
+            id="canny_high_threshold",
+        )
+        debug_canny: bool = Field(
+            False,
+            title="Debug Canny",
+            field="checkbox",
+            hide=True,
+            id="debug_canny",
+        )
+
+    def __init__(self, args: Args, device: torch.device, torch_dtype: torch.dtype):
+        controlnet_canny = ControlNetModel.from_pretrained(
+            controlnet_model, torch_dtype=torch_dtype
+        ).to(device)
+        vae = AutoencoderKL.from_pretrained(
+            "madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch_dtype
+        )
+        if args.safety_checker:
+            self.pipe = StableDiffusionXLControlNetImg2ImgPipeline.from_pretrained(
+                model_id,
+                controlnet=controlnet_canny,
+                vae=vae,
+            )
+        else:
+            self.pipe = StableDiffusionXLControlNetImg2ImgPipeline.from_pretrained(
+                model_id,
+                safety_checker=None,
+                controlnet=controlnet_canny,
+                vae=vae,
+            )
+        self.canny_torch = SobelOperator(device=device)
+
+        self.pipe.set_progress_bar_config(disable=True)
+        self.pipe.to(device=device, dtype=torch_dtype).to(device)
+        if device.type != "mps":
+            self.pipe.unet.to(memory_format=torch.channels_last)
+
+        if psutil.virtual_memory().total < 64 * 1024**3:
+            self.pipe.enable_attention_slicing()
+
+        self.pipe.compel_proc = Compel(
+            tokenizer=[self.pipe.tokenizer, self.pipe.tokenizer_2],
+            text_encoder=[self.pipe.text_encoder, self.pipe.text_encoder_2],
+            returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
+            requires_pooled=[False, True],
+        )
+        if args.use_taesd:
+            self.pipe.vae = AutoencoderTiny.from_pretrained(
+                taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+            ).to(device)
+
+        if args.torch_compile:
+            self.pipe.unet = torch.compile(
+                self.pipe.unet, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe.vae = torch.compile(
+                self.pipe.vae, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe(
+                prompt="warmup",
+                image=[Image.new("RGB", (768, 768))],
+                control_image=[Image.new("RGB", (768, 768))],
+            )
+
+    def predict(self, params: "Pipeline.InputParams") -> Image.Image:
+        generator = torch.manual_seed(params.seed)
+
+        prompt_embeds, pooled_prompt_embeds = self.pipe.compel_proc(
+            [params.prompt, params.negative_prompt]
+        )
+        control_image = self.canny_torch(
+            params.image, params.canny_low_threshold, params.canny_high_threshold
+        )
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
+
+        results = self.pipe(
+            image=params.image,
+            control_image=control_image,
+            prompt_embeds=prompt_embeds[0:1],
+            pooled_prompt_embeds=pooled_prompt_embeds[0:1],
+            negative_prompt_embeds=prompt_embeds[1:2],
+            negative_pooled_prompt_embeds=pooled_prompt_embeds[1:2],
+            generator=generator,
+            strength=strength,
+            num_inference_steps=steps,
+            guidance_scale=params.guidance_scale,
+            width=params.width,
+            height=params.height,
+            output_type="pil",
+            controlnet_conditioning_scale=params.controlnet_scale,
+            control_guidance_start=params.controlnet_start,
+            control_guidance_end=params.controlnet_end,
+        )
+
+        nsfw_content_detected = (
+            results.nsfw_content_detected[0]
+            if "nsfw_content_detected" in results
+            else False
+        )
+        if nsfw_content_detected:
+            return None
+        result_image = results.images[0]
+        if params.debug_canny:
+            # paste control_image on top of result_image
+            w0, h0 = (200, 200)
+            control_image = control_image.resize((w0, h0))
+            w1, h1 = result_image.size
+            result_image.paste(control_image, (w1 - w0, h1 - h0))
+
+        return result_image

pipelines/img2img.py

@@ -14,11 +14,36 @@ import psutil
 from config import Args
 from pydantic import BaseModel, Field
 from PIL import Image
+import math
 
 base_model = "SimianLuo/LCM_Dreamshaper_v7"
 taesd_model = "madebyollin/taesd"
 
 default_prompt = "Portrait of The Terminator with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model</h1>
+<h3 class="text-xl font-bold">Image-to-Image LCM</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/blog/lcm_lora"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM</a>
+Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
 
 
 class Pipeline:
@@ -27,6 +52,7 @@ class Pipeline:
         title: str = "Image-to-Image LCM"
         description: str = "Generates an image from a text prompt"
         input_mode: str = "image"
+        page_content: str = page_content
 
     class InputParams(BaseModel):
         prompt: str = Field(
@@ -39,13 +65,13 @@ class Pipeline:
             2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
         )
         steps: int = Field(
-            4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
+            4, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
-            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+            768, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
         )
         height: int = Field(
-            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+            768, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
         )
         guidance_scale: float = Field(
             0.2,
@@ -79,7 +105,7 @@ class Pipeline:
         if args.use_taesd:
             self.pipe.vae = AutoencoderTiny.from_pretrained(
                 taesd_model, torch_dtype=torch_dtype, use_safetensors=True
-            )
+            ).to(device)
 
         self.pipe.set_progress_bar_config(disable=True)
         self.pipe.to(device=device, dtype=torch_dtype)
@@ -113,12 +139,18 @@ class Pipeline:
     def predict(self, params: "Pipeline.InputParams") -> Image.Image:
         generator = torch.manual_seed(params.seed)
         prompt_embeds = self.compel_proc(params.prompt)
+
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
+
         results = self.pipe(
             image=params.image,
             prompt_embeds=prompt_embeds,
             generator=generator,
-            strength=params.strength,
-            num_inference_steps=params.steps,
+            strength=strength,
+            num_inference_steps=steps,
             guidance_scale=params.guidance_scale,
             width=params.width,
             height=params.height,

pipelines/img2imgSDXLTurbo.py

@@ -0,0 +1,182 @@
+from diffusers import (
+    AutoPipelineForImage2Image,
+    AutoencoderTiny,
+)
+from compel import Compel, ReturnedEmbeddingsType
+import torch
+
+try:
+    import intel_extension_for_pytorch as ipex  # type: ignore
+except:
+    pass
+
+import psutil
+from config import Args
+from pydantic import BaseModel, Field
+from PIL import Image
+import math
+
+base_model = "stabilityai/sdxl-turbo"
+taesd_model = "madebyollin/taesdxl"
+
+default_prompt = "close-up photography of old man standing in the rain at night, in a street lit by lamps, leica 35mm summilux"
+default_negative_prompt = "blurry, low quality, render, 3D, oversaturated"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time SDXL Turbo</h1>
+<h3 class="text-xl font-bold">Image-to-Image</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/stabilityai/sdxl-turbo"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">SDXL Turbo</a>
+Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/sdxl_turbo"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
+
+
+class Pipeline:
+    class Info(BaseModel):
+        name: str = "img2img"
+        title: str = "Image-to-Image SDXL"
+        description: str = "Generates an image from a text prompt"
+        input_mode: str = "image"
+        page_content: str = page_content
+
+    class InputParams(BaseModel):
+        prompt: str = Field(
+            default_prompt,
+            title="Prompt",
+            field="textarea",
+            id="prompt",
+        )
+        negative_prompt: str = Field(
+            default_negative_prompt,
+            title="Negative Prompt",
+            field="textarea",
+            id="negative_prompt",
+            hide=True,
+        )
+        seed: int = Field(
+            2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
+        )
+        steps: int = Field(
+            4, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
+        )
+        width: int = Field(
+            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+        )
+        height: int = Field(
+            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+        )
+        guidance_scale: float = Field(
+            0.2,
+            min=0,
+            max=20,
+            step=0.001,
+            title="Guidance Scale",
+            field="range",
+            hide=True,
+            id="guidance_scale",
+        )
+        strength: float = Field(
+            0.5,
+            min=0.25,
+            max=1.0,
+            step=0.001,
+            title="Strength",
+            field="range",
+            hide=True,
+            id="strength",
+        )
+
+    def __init__(self, args: Args, device: torch.device, torch_dtype: torch.dtype):
+        if args.safety_checker:
+            self.pipe = AutoPipelineForImage2Image.from_pretrained(base_model)
+        else:
+            self.pipe = AutoPipelineForImage2Image.from_pretrained(
+                base_model,
+                safety_checker=None,
+            )
+        if args.use_taesd:
+            self.pipe.vae = AutoencoderTiny.from_pretrained(
+                taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+            ).to(device)
+
+        self.pipe.set_progress_bar_config(disable=True)
+        self.pipe.to(device=device, dtype=torch_dtype)
+        if device.type != "mps":
+            self.pipe.unet.to(memory_format=torch.channels_last)
+
+        # check if computer has less than 64GB of RAM using sys or os
+        if psutil.virtual_memory().total < 64 * 1024**3:
+            self.pipe.enable_attention_slicing()
+
+        if args.torch_compile:
+            print("Running torch compile")
+            self.pipe.unet = torch.compile(
+                self.pipe.unet, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe.vae = torch.compile(
+                self.pipe.vae, mode="reduce-overhead", fullgraph=True
+            )
+
+            self.pipe(
+                prompt="warmup",
+                image=[Image.new("RGB", (768, 768))],
+            )
+
+        self.pipe.compel_proc = Compel(
+            tokenizer=[self.pipe.tokenizer, self.pipe.tokenizer_2],
+            text_encoder=[self.pipe.text_encoder, self.pipe.text_encoder_2],
+            returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
+            requires_pooled=[False, True],
+        )
+
+    def predict(self, params: "Pipeline.InputParams") -> Image.Image:
+        generator = torch.manual_seed(params.seed)
+        prompt_embeds, pooled_prompt_embeds = self.pipe.compel_proc(
+            [params.prompt, params.negative_prompt]
+        )
+        steps = params.steps
+        strength = params.strength
+        if int(steps * strength) < 1:
+            steps = math.ceil(1 / max(0.10, strength))
+
+        results = self.pipe(
+            image=params.image,
+            prompt_embeds=prompt_embeds[0:1],
+            pooled_prompt_embeds=pooled_prompt_embeds[0:1],
+            negative_prompt_embeds=prompt_embeds[1:2],
+            negative_pooled_prompt_embeds=pooled_prompt_embeds[1:2],
+            generator=generator,
+            strength=strength,
+            num_inference_steps=steps,
+            guidance_scale=params.guidance_scale,
+            width=params.width,
+            height=params.height,
+            output_type="pil",
+        )
+
+        nsfw_content_detected = (
+            results.nsfw_content_detected[0]
+            if "nsfw_content_detected" in results
+            else False
+        )
+        if nsfw_content_detected:
+            return None
+        result_image = results.images[0]
+
+        return result_image

pipelines/txt2img.py

@@ -17,6 +17,28 @@ taesd_model = "madebyollin/taesd"
 
 default_prompt = "Portrait of The Terminator with , glare pose, detailed, intricate, full of colour, cinematic lighting, trending on artstation, 8k, hyperrealistic, focused, extreme details, unreal engine 5 cinematic, masterpiece"
 
+page_content = """<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model</h1>
+<h3 class="text-xl font-bold">Text-to-Image</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/SimianLuo/LCM_Dreamshaper_v7"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM</a>
+Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a> with a MJPEG stream server
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>"""
+
 
 class Pipeline:
     class Info(BaseModel):
@@ -24,6 +46,7 @@ class Pipeline:
         title: str = "Text-to-Image LCM"
         description: str = "Generates an image from a text prompt"
         input_mode: str = "text"
+        page_content: str = page_content
 
     class InputParams(BaseModel):
         prompt: str = Field(
@@ -39,10 +62,10 @@ class Pipeline:
             4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
-            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+            768, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
         )
         height: int = Field(
-            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+            768, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
         )
         guidance_scale: float = Field(
             8.0,
@@ -65,7 +88,7 @@ class Pipeline:
         if args.use_taesd:
             self.pipe.vae = AutoencoderTiny.from_pretrained(
                 taesd_model, torch_dtype=torch_dtype, use_safetensors=True
-            )
+            ).to(device)
 
         self.pipe.set_progress_bar_config(disable=True)
         self.pipe.to(device=device, dtype=torch_dtype)

pipelines/txt2imgLora.py

@@ -18,6 +18,34 @@ taesd_model = "madebyollin/taesd"
 
 default_prompt = "Analog style photograph of young Harrison Ford as Han Solo, star wars behind the scenes"
 
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model SDv1.5</h1>
+<h3 class="text-xl font-bold">Text-to-Image LCM + LoRa</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/blog/lcm_lora"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM</a>
+Image to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server. Featuring <a
+    href="https://huggingface.co/wavymulder/Analog-Diffusion"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Analog-Diffusion</a>
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
+
 
 class Pipeline:
     class Info(BaseModel):
@@ -25,6 +53,7 @@ class Pipeline:
         title: str = "Text-to-Image LCM + LoRa"
         description: str = "Generates an image from a text prompt"
         input_mode: str = "text"
+        page_content: str = page_content
 
     class InputParams(BaseModel):
         prompt: str = Field(
@@ -66,7 +95,7 @@ class Pipeline:
         if args.use_taesd:
             self.pipe.vae = AutoencoderTiny.from_pretrained(
                 taesd_model, torch_dtype=torch_dtype, use_safetensors=True
-            )
+            ).to(device)
         self.pipe.scheduler = LCMScheduler.from_config(self.pipe.scheduler.config)
         self.pipe.set_progress_bar_config(disable=True)
         self.pipe.to(device=device, dtype=torch_dtype)

pipelines/txt2imgLoraSDXL.py

@@ -1,8 +1,4 @@
-from diffusers import (
-    DiffusionPipeline,
-    LCMScheduler,
-    AutoencoderKL,
-)
+from diffusers import DiffusionPipeline, LCMScheduler, AutoencoderKL, AutoencoderTiny
 from compel import Compel, ReturnedEmbeddingsType
 import torch
 
@@ -16,13 +12,38 @@ from config import Args
 from pydantic import BaseModel, Field
 from PIL import Image
 
-controlnet_model = "diffusers/controlnet-canny-sdxl-1.0"
 model_id = "stabilityai/stable-diffusion-xl-base-1.0"
 lcm_lora_id = "latent-consistency/lcm-lora-sdxl"
+taesd_model = "madebyollin/taesdxl"
 
 
 default_prompt = "close-up photography of old man standing in the rain at night, in a street lit by lamps, leica 35mm summilux"
 default_negative_prompt = "blurry, low quality, render, 3D, oversaturated"
+page_content = """
+<h1 class="text-3xl font-bold">Real-Time Latent Consistency Model</h1>
+<h3 class="text-xl font-bold">Text-to-Image SDXL + LCM + LoRA</h3>
+<p class="text-sm">
+    This demo showcases
+    <a
+    href="https://huggingface.co/blog/lcm_lora"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">LCM LoRA</a
+    >
+    Text to Image pipeline using
+    <a
+    href="https://huggingface.co/docs/diffusers/main/en/using-diffusers/lcm#performing-inference-with-lcm"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Diffusers</a
+    > with a MJPEG stream server.
+</p>
+<p class="text-sm text-gray-500">
+    Change the prompt to generate different images, accepts <a
+    href="https://github.com/damian0815/compel/blob/main/doc/syntax.md"
+    target="_blank"
+    class="text-blue-500 underline hover:no-underline">Compel</a
+    > syntax.
+</p>
+"""
 
 
 class Pipeline:
@@ -30,6 +51,7 @@ class Pipeline:
         name: str = "LCM+Lora+SDXL"
         title: str = "Text-to-Image SDXL + LCM + LoRA"
         description: str = "Generates an image from a text prompt"
+        page_content: str = page_content
         input_mode: str = "text"
 
     class InputParams(BaseModel):
@@ -50,7 +72,7 @@ class Pipeline:
             2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
         )
         steps: int = Field(
-            4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
+            4, min=1, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
             1024, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
@@ -101,6 +123,10 @@ class Pipeline:
             returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
             requires_pooled=[False, True],
         )
+        if args.use_taesd:
+            self.pipe.vae = AutoencoderTiny.from_pretrained(
+                taesd_model, torch_dtype=torch_dtype, use_safetensors=True
+            ).to(device)
 
         if args.torch_compile:
             self.pipe.unet = torch.compile(

requirements.txt

@@ -1,4 +1,4 @@
-git+https://github.com/huggingface/diffusers@6f1435332bc74e286af5e88014236e4cc712b747
+git+https://github.com/huggingface/diffusers@dadd55fb36acc862254cf935826d54349b0fcd8c
 transformers==4.35.2
 --extra-index-url https://download.pytorch.org/whl/cu121;
 torch==2.1.0
@@ -9,4 +9,5 @@ accelerate==0.24.0
 compel==2.0.2
 controlnet-aux==0.0.7
 peft==0.6.0
-xformers; sys_platform != 'darwin' or platform_machine != 'arm64'
+xformers; sys_platform != 'darwin' or platform_machine != 'arm64'
+markdown2