Update app.py

app.py

@@ -837,13 +837,37 @@ def on_image_upload(image, video, high_res):
     return gr.update(value=w), gr.update(value=h)
 
 
+def _get_video_duration(video_path) -> float | None:
+    """Get video duration in seconds via ffprobe."""
+    if video_path is None:
+        return None
+    try:
+        result = subprocess.run(
+            ["ffprobe", "-v", "error", "-select_streams", "v:0",
+             "-show_entries", "format=duration", "-of", "default=nw=1:nk=1",
+             str(video_path)],
+            capture_output=True, text=True,
+        )
+        return float(result.stdout.strip())
+    except Exception:
+        return None
+
+
 def on_video_upload(video, image, high_res):
-    """Auto-set resolution when video is uploaded."""
+    """Auto-set resolution and duration when video is uploaded."""
     media = video if video is not None else image
     aspect = detect_aspect_ratio(media)
     tier = "high" if high_res else "low"
     w, h = RESOLUTIONS[tier][aspect]
-    return gr.update(value=w), gr.update(value=h)
+
+    # Auto-adjust duration to min(video_length, 10)
+    vid_dur = _get_video_duration(video)
+    if vid_dur is not None:
+        dur = round(min(vid_dur, 10.0), 1)
+    else:
+        dur = 3.0
+
+    return gr.update(value=w), gr.update(value=h), gr.update(value=dur)
 
 
 def on_highres_toggle(image, video, high_res):
@@ -999,87 +1023,82 @@ def generate_video(
 
 
 # ─────────────────────────────────────────────────────────────────────────────
-# Gradio UI
+# Gradio UI — LTX 2.3 Move
 # ─────────────────────────────────────────────────────────────────────────────
-with gr.Blocks(title="LTX-2.3 Unified: V2V + I2V + A2V") as demo:
-    gr.Markdown("# LTX-2.3 Unified: Video/Image/Audio → Video")
-    gr.Markdown(
-        "Unified pipeline for **video-to-video** (IC-LoRA), **image-to-video**, "
-        "and **audio-conditioned** generation with LTX-2.3 — use any combination of inputs. "
-        "[[model]](https://huggingface.co/Lightricks/LTX-2.3) "
-        "[[code]](https://github.com/Lightricks/LTX-2)"
-    )
+css = """
+.main-title { text-align: center; margin-bottom: 0.5em; }
+.generate-btn { min-height: 52px !important; font-size: 1.1em !important; }
+footer { display: none !important; }
+"""
+
+purple_citrus = gr.themes.Citrus(
+    primary_hue=gr.themes.colors.purple,
+    secondary_hue=gr.themes.colors.purple,
+    neutral_hue=gr.themes.colors.gray,
+)
+
+with gr.Blocks(title="LTX 2.3 Move") as demo:
+    gr.Markdown("<h1 class='main-title'>LTX 2.3 Move</h1>")
 
-    with gr.Row():
-        with gr.Column():
-            # All three inputs visible at once
+    # Hidden state — preprocessing is always Pose
+    video_preprocess = gr.State("Pose (DWPose)")
+
+    with gr.Row(equal_height=True):
+        # ── Left column: inputs ──────────────────────────────────────
+        with gr.Column(scale=1):
             with gr.Row():
+                input_video = gr.Video(
+                    label="Motion Source (video)",
+                    sources=["upload"],
+                )
                 input_image = gr.Image(
-                    label="🖼️ Input Image (I2V — first frame)",
+                    label="Style Source (image, optional)",
                     type="filepath",
                 )
-                with gr.Column():
-                    input_video = gr.Video(
-                        label="🎬 Reference Video (V2V)",
-                        sources=["upload"],
-                    )
-                    video_preprocess = gr.Dropdown(
-                        label="Video Preprocessing",
-                        choices=[
-                            "Pose (DWPose)",
-                            "Canny Edge",
-                            "Depth (Laplacian)",
-                            "Raw (no preprocessing)",
-                        ],
-                        value="Pose (DWPose)",
-                        info="Strips appearance from video → style comes from image/prompt instead",
-                    )
-            input_audio = gr.Audio(
-                label="🔊 Input Audio (A2V — lipsync / BGM)",
-                type="filepath",
-            )
 
             prompt = gr.Textbox(
                 label="Prompt",
-                info="Describe the desired output — be as detailed as possible",
                 value="Make this come alive with cinematic motion, smooth animation",
-                lines=3,
-                placeholder="Describe the motion, style, and content you want...",
+                lines=2,
+                placeholder="Describe the style, motion, and content you want...",
             )
 
             with gr.Row():
                 duration = gr.Slider(
-                    label="Duration (seconds)",
-                    minimum=1.0, maximum=10.0, value=3.0, step=0.1,
+                    label="Duration (s)", minimum=1.0, maximum=10.0, value=3.0, step=0.1,
                 )
+                enhance_prompt = gr.Checkbox(label="Enhance Prompt", value=True)
+
+            generate_btn = gr.Button(
+                "Generate", variant="primary", size="lg", elem_classes=["generate-btn"],
+            )
+
+            with gr.Accordion("Advanced Settings", open=False):
                 conditioning_strength = gr.Slider(
                     label="V2V Conditioning Strength",
-                    info="How closely to follow the reference video",
-                    minimum=0.0, maximum=1.0, value=1.0, step=0.05,
+                    info="How closely to follow the reference video's structure",
+                    minimum=0.0, maximum=1.0, value=0.85, step=0.05,
                 )
-
-            with gr.Row():
-                enhance_prompt = gr.Checkbox(label="Enhance Prompt", value=True)
-                high_res = gr.Checkbox(label="High Resolution", value=True)
+                high_res = gr.Checkbox(label="High Resolution (2×)", value=False)
                 use_video_audio = gr.Checkbox(
-                    label="Use Audio from Video",
-                    value=True,
-                    info="Extract and use the audio track from the reference video",
+                    label="Use Audio from Video", value=True,
+                    info="Extract the audio track from the motion source video",
+                )
+                input_audio = gr.Audio(
+                    label="Override Audio (optional — replaces video audio)",
+                    type="filepath",
                 )
-
-            generate_btn = gr.Button("Generate Video", variant="primary", size="lg")
-
-            with gr.Accordion("Advanced Settings", open=False):
                 seed = gr.Slider(
                     label="Seed", minimum=0, maximum=MAX_SEED, value=42, step=1,
                 )
                 randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
                 with gr.Row():
-                    width = gr.Number(label="Width", value=1536, precision=0)
-                    height = gr.Number(label="Height", value=1024, precision=0)
+                    width = gr.Number(label="Width", value=768, precision=0)
+                    height = gr.Number(label="Height", value=512, precision=0)
 
-        with gr.Column():
-            output_video = gr.Video(label="Generated Video", autoplay=True)
+        # ── Right column: output ─────────────────────────────────────
+        with gr.Column(scale=1):
+            output_video = gr.Video(label="Result", autoplay=True)
 
     # ── Event handlers ───────────────────────────────────────────────────
     input_image.change(
@@ -1090,7 +1109,7 @@ with gr.Blocks(title="LTX-2.3 Unified: V2V + I2V + A2V") as demo:
     input_video.change(
         fn=on_video_upload,
         inputs=[input_video, input_image, high_res],
-        outputs=[width, height],
+        outputs=[width, height, duration],
     )
     high_res.change(
         fn=on_highres_toggle,
@@ -1109,4 +1128,4 @@ with gr.Blocks(title="LTX-2.3 Unified: V2V + I2V + A2V") as demo:
 
 
 if __name__ == "__main__":
-    demo.launch(theme=gr.themes.Citrus())
+    demo.launch(css=css, theme=purple_citrus)