Fix: pass image through chat template content so pixel_values are produced

app.py

@@ -19,21 +19,6 @@ model = AutoModelForImageTextToText.from_pretrained(
 print("Model loaded!")
 
 
-def _load_image(image):
-    """Accept a workflow image value (dict with 'path', or a path/URL string)
-    and return a PIL.Image."""
-    from PIL import Image
-
-    if isinstance(image, dict):
-        image = image.get("path") or image.get("url")
-    if isinstance(image, str) and image.startswith(("http://", "https://")):
-        import requests
-        from io import BytesIO
-
-        return Image.open(BytesIO(requests.get(image, timeout=30).content)).convert("RGB")
-    return Image.open(image).convert("RGB")
-
-
 def _estimate_duration(image, prompt, max_new_tokens, temperature, top_p, top_k) -> int:
     """Rough wall-clock estimate (seconds) for one VLM call. Requesting less
     than the 60s default raises queue priority and frees the GPU slot sooner
@@ -70,13 +55,16 @@ def _run_vlm_gpu(image, prompt, max_new_tokens, temperature, top_p, top_k):
     if image is None:
         raise gr.Error("Please provide an image.")
 
-    pil_image = _load_image(image)
+    if isinstance(image, dict):
+        image_ref = image.get("path") or image.get("url")
+    else:
+        image_ref = image
 
     messages = [
         {
             "role": "user",
             "content": [
-                {"type": "image"},
+                {"type": "image", "url": image_ref},
                 {"type": "text", "text": prompt},
             ],
         }
@@ -88,9 +76,9 @@ def _run_vlm_gpu(image, prompt, max_new_tokens, temperature, top_p, top_k):
         add_generation_prompt=True,
         return_tensors="pt",
         return_dict=True,
-    )
-    inputs = {k: v.to(model.device) if hasattr(v, "to") else v for k, v in inputs.items()}
-    inputs["pixel_values"] = inputs["pixel_values"].to(torch.bfloat16)
+    ).to(model.device)
+    if "pixel_values" in inputs:
+        inputs["pixel_values"] = inputs["pixel_values"].to(torch.bfloat16)
 
     do_sample = float(temperature) > 0
     gen_kwargs = dict(max_new_tokens=int(max_new_tokens), do_sample=do_sample)