Return serveable file dicts from fn nodes; enable hf_oauth

call_fn JSON-serializes bound-function results verbatim (unlike call_space,
which rewrites local paths into {path,url,is_file} dicts), so the generated
mp4 never rendered on the canvas. Wrap the video output in the file-dict
shape with a /gradio_api/file= URL; the tempdir is already in allowed_paths.

Also add hf_oauth: true so visitors run workflows on their own inference
quota and the owner can edit the canvas.

README.md

@@ -5,6 +5,7 @@ colorFrom: gray
 colorTo: red
 sdk: gradio
 sdk_version: 6.28.0
+hf_oauth: true
 app_file: app.py
 python_version: "3.12"
 startup_duration_timeout: 1h

app.py

@@ -434,22 +434,42 @@ def _as_path(value):
     return value
 
 
+def _as_file(value):
+    """`call_fn` JSON-serializes bound-function results verbatim (only `call_space`
+    rewrites local paths into serveable file dicts), so media outputs must be returned
+    in the `{path, url, is_file}` shape the canvas renders. The video lives under the
+    system tempdir, which `Workflow.launch()` adds to `allowed_paths`."""
+    if not isinstance(value, str) or not os.path.exists(value):
+        return value
+    try:
+        from gradio_client import utils as client_utils
+
+        encoded = client_utils.encode_file_path(value)
+    except (ImportError, AttributeError):
+        import urllib.parse
+
+        encoded = urllib.parse.quote(os.path.abspath(value))
+    return {"path": value, "url": "/gradio_api/file=" + encoded, "is_file": True}
+
+
 def i2v(image, prompt, actions, negative_prompt=NEGATIVE_PROMPT,
         num_chunks=DEFAULT_CHUNKS, seed=42):
     """Image → video node: explore an uploaded scene with a scripted camera."""
-    return generate_i2v(
+    video, info = generate_i2v(
         _as_path(image), prompt, actions, negative_prompt or NEGATIVE_PROMPT,
         int(num_chunks), int(seed),
     )
+    return _as_file(video), info
 
 
 def t2v(prompt, actions, negative_prompt=NEGATIVE_PROMPT,
         num_chunks=DEFAULT_CHUNKS, seed=42):
     """Text → video node: generate a world from text and explore it."""
-    return generate_t2v(
+    video, info = generate_t2v(
         prompt, actions, negative_prompt or NEGATIVE_PROMPT,
         int(num_chunks), int(seed),
     )
+    return _as_file(video), info
 
 
 CSS = """