Update app.py

app.py

@@ -1,36 +1,73 @@
 import gradio as gr
 from PIL import Image
-import requests
+import yt_dlp
+import cv2
+import tempfile
+import time
 from transformers import BlipProcessor, BlipForConditionalGeneration
-from io import BytesIO
 
 processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
 model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
 
-def analyse_youtube_image(youtube_url):
-    # Pak thumbnail van YouTube als beeld
-    try:
-        video_id = youtube_url.split("v=")[-1]
-        thumb_url = f"https://img.youtube.com/vi/{video_id}/maxresdefault.jpg"
-        response = requests.get(thumb_url)
-        image = Image.open(BytesIO(response.content)).convert("RGB")
-    except Exception as e:
-        return f"Kon geen frame laden: {e}", None
-
-    # Caption het beeld
-    inputs = processor(images=image, return_tensors="pt")
-    out = model.generate(**inputs)
-    caption = processor.decode(out[0], skip_special_tokens=True)
-    return caption, image
+def get_video_frame(youtube_url, seek_time=0):
+    # Download een stukje van de video (paar seconden), pak een frame rond seek_time
+    with tempfile.NamedTemporaryFile(suffix=".mp4") as temp_video:
+        ydl_opts = {
+            'outtmpl': temp_video.name,
+            'format': 'bestvideo[ext=mp4]+bestaudio[ext=m4a]/mp4',
+            'quiet': True,
+            'noplaylist': True,
+            'download_ranges': f"*{seek_time}-{seek_time+1}",
+            'retries': 3,
+        }
+        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+            try:
+                ydl.download([youtube_url])
+            except Exception as e:
+                return None, f"Download error: {str(e)}"
+
+        # Pak frame
+        vidcap = cv2.VideoCapture(temp_video.name)
+        vidcap.set(cv2.CAP_PROP_POS_MSEC, 500)  # Pak frame halverwege het stukje
+        success, image = vidcap.read()
+        vidcap.release()
+        if success:
+            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+            pil_image = Image.fromarray(image)
+            return pil_image, None
+        else:
+            return None, "Kon geen frame uitlezen."
+
+def analyse_stream(youtube_url, interval=10, num_frames=3):
+    results = []
+    for i in range(num_frames):
+        seek = i * interval
+        img, err = get_video_frame(youtube_url, seek)
+        if err or img is None:
+            results.append((f"Fout: {err}", None))
+            continue
+        # Caption
+        inputs = processor(images=img, return_tensors="pt")
+        out = model.generate(**inputs)
+        caption = processor.decode(out[0], skip_special_tokens=True)
+        results.append((caption, img))
+    return results
+
+def gradio_multi(youtube_url):
+    res = analyse_stream(youtube_url, interval=10, num_frames=3)
+    texts = [r[0] for r in res]
+    imgs = [r[1] for r in res]
+    return texts, imgs
 
 with gr.Blocks() as demo:
-    gr.Markdown("# YouTube Thumbnail Analyse met BLIP\nVoer een YouTube-link in. Je krijgt een beschrijving van het beeld.")
+    gr.Markdown("# 🎥 YouTube livestream analyse (meerdere frames)")
     youtube_url = gr.Textbox(label="YouTube URL", value="https://www.youtube.com/watch?v=R5i7aeV8SB8")
-    output = gr.Textbox(label="Model antwoord")
-    image_out = gr.Image(label="Thumbnail")
-    run_btn = gr.Button("Analyseer!")
+    run_btn = gr.Button("Analyseer 3 beelden (om de 10 sec)")
+    output = gr.Dataframe(label="Model antwoorden", headers=["Beschrijving"])
+    images = gr.Gallery(label="Frames")
 
-    run_btn.click(analyse_youtube_image, inputs=youtube_url, outputs=[output, image_out])
+    run_btn.click(gradio_multi, inputs=youtube_url, outputs=[output, images])
 
 demo.launch()
 
+