[Model] Migrate MammothModa2-Preview example to standard task example + model_extras (#4691)

Signed-off-by: mo saif <moisf56@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

examples/offline_inference/text_to_image/text_to_image.py

@@ -589,6 +589,12 @@ def main():
         if images:
             break
 
+    # Fallback: generation-stage pipelines (e.g. MammothModa2's AR->DiT) return the
+    # generated image as a tensor under multimodal_output instead of populating the
+    # `images` field that diffusion-stage pipelines fill.
+    if not images:
+        images = _images_from_multimodal_output(outputs)
+
     if not images:
         raise ValueError("No images found in request_output")
 
@@ -606,5 +612,36 @@ def main():
             print(f"Saved generated image to {save_path}")
 
 
+def _images_from_multimodal_output(outputs: list[Any]) -> list[Any]:
+    """Extract PIL images from multimodal_output tensors.
+
+    Generation-stage pipelines (e.g. MammothModa2's AR->DiT) return the generated
+    image as a tensor (normalized to [-1, 1], CHW) under ``multimodal_output``
+    rather than populating the ``images`` field. Convert any such tensors to PIL.
+    """
+    from PIL import Image
+
+    pil_images: list[Any] = []
+    for output in outputs:
+        req_out = getattr(output, "request_output", output)
+        for completion in getattr(req_out, "outputs", None) or []:
+            # multimodal_output is a MultimodalPayload (a Mapping) keyed by modality,
+            # matching how omni examples (ming_flash_omni / magi_human / dynin) read it.
+            mm = getattr(completion, "multimodal_output", None) or {}
+            if "image" not in mm:
+                continue
+            payload = mm["image"]
+            for tensor in payload if isinstance(payload, list) else [payload]:
+                if not isinstance(tensor, torch.Tensor):
+                    continue
+                img = tensor.detach().to("cpu", dtype=torch.float32)
+                if img.ndim == 4:
+                    img = img[0]
+                img = (img / 2 + 0.5).clamp(0, 1).mul(255).to(torch.uint8)
+                img = img.permute(1, 2, 0).contiguous().numpy()
+                pil_images.append(Image.fromarray(img))
+    return pil_images
+
+
 if __name__ == "__main__":
     main()