feat(app): add auto-canvas selection and update conditioner space

- Update `CONDITIONER_SPACE` default to `mpasila/qwen3vl-conditioner`.
- Add 3:2 and 2:3 aspect ratio options to `CANVASES`.
- Implement `auto_canvas` logic to automatically select the closest canvas aspect ratio based on uploaded image or video dimensions.
- Bind `auto_canvas` to image and video component change events in the Gradio UI.

app.py

@@ -20,7 +20,7 @@ import spaces
 import gradio as gr
 
 MODEL_REPO = os.environ.get("H3_MODEL_REPO", "MiniMaxAI/MiniMax-H3")
-CONDITIONER_SPACE = os.environ.get("H3_CONDITIONER", "multimodalart/qwen3vl-conditioner")
+CONDITIONER_SPACE = os.environ.get("H3_CONDITIONER", "mpasila/qwen3vl-conditioner")
 # `lazy` moves all 72.16 GiB onto the card on the first GPU call and leaves it there; `offload` hands placement to
 # `ComponentsManager.enable_auto_cpu_offload`. Startup placement is not an option here — see `load_models`.
 PLACEMENT = os.environ.get("H3_PLACEMENT", "lazy").lower()
@@ -54,6 +54,11 @@ CANVASES = {
     "1024x768 · 4:3 full": (768, 1024),
     "576x768 · 3:4 fast": (768, 576),
     "768x1024 · 3:4 full": (1024, 768),
+    # 3:2 / 2:3
+    "864x576 · 3:2 fast": (576, 864),
+    "1152x768 · 3:2 full": (768, 1152),
+    "576x864 · 2:3 fast": (864, 576),
+    "768x1152 · 2:3 full": (1152, 768),
     # 21:9
     "1152x512 · 21:9 fast": (512, 1152),
     "1536x672 · 21:9 full": (672, 1536),
@@ -498,6 +503,43 @@ def probe(path: str) -> tuple[float | None, float | None]:
     return video, audio
 
 
+def _media_dimensions(path: str) -> tuple[int, int]:
+    """`(width, height)` of an image or a video file, from its first stream."""
+    from PIL import Image
+
+    try:
+        with Image.open(path) as image:
+            return image.size
+    except Exception:
+        pass
+    import av
+
+    with av.open(path) as container:
+        stream = container.streams.video[0]
+        return stream.width, stream.height
+
+
+def closest_canvas(path: str | None) -> str | None:
+    """The canvas label whose aspect ratio is closest to a media file's, or `None` when the file is
+    missing or unreadable."""
+    if not path:
+        return None
+    try:
+        width, height = _media_dimensions(path)
+    except Exception:
+        return None
+    if not width or not height:
+        return None
+    target = width / height
+    return min(CANVASES, key=lambda label: abs(CANVASES[label][1] / CANVASES[label][0] - target))
+
+
+def auto_canvas(path):
+    """Set the canvas to the closest aspect ratio of an uploaded image or video."""
+    label = closest_canvas(path)
+    return gr.update(value=label) if label else gr.update()
+
+
 def collect(image_paths, audio_path, video_path) -> list[tuple[str, str]]:
     """The `(kind, path)` references of a request, **in the order the model reads them**.
 
@@ -947,6 +989,11 @@ with gr.Blocks(title="MiniMax-H3 Reference Custom Lora") as demo:
             duration_controls, [audio, video, match], [match, duration], show_progress="hidden", api_name=False
         )
 
+    # Auto-select the canvas whose aspect ratio is closest to an uploaded image or video.
+    for image in images:
+        image.change(auto_canvas, image, canvas, show_progress="hidden", api_name=False)
+    video.change(auto_canvas, video, canvas, show_progress="hidden", api_name=False)
+
     # `reference, strength, reference, strength, ...`, which is how `generate` unpacks them.
     lora_inputs = [field for pair in zip(lora_references, lora_scales) for field in pair]
     lora_upload.upload(_fill_lora_slots, [lora_upload, *lora_references], lora_references, api_name=False)