audio folded into single prompt; inpaint example None->placeholder mask

app.py

@@ -44,11 +44,8 @@ pipe.set_adapters("refsheet", LORA_SCALE)
 
 
 # --- Helpers ----------------------------------------------------------------
-def _build_prompt(sheet, action, audio):
-    p = f"Reference sheet: {sheet.strip()}\n\nGenerated video: {action.strip()}"
-    if audio.strip():
-        p += f" Audio: {audio.strip()}."
-    return p
+def _build_prompt(sheet, action):
+    return f"Reference sheet: {sheet.strip()}\n\nGenerated video: {action.strip()}"
 
 
 def _export(video_np, audio, path):
@@ -64,7 +61,7 @@ def _duration(*args, **kwargs):
 
 # --- Inference --------------------------------------------------------------
 @spaces.GPU(duration=_duration)
-def generate(image, sheet, action, audio, lora_scale, seed, randomize,
+def generate(image, sheet, action, lora_scale, seed, randomize,
              progress=gr.Progress(track_tqdm=True)):
     if image is None:
         raise gr.Error("Please upload a reference sheet image.")
@@ -80,7 +77,7 @@ def generate(image, sheet, action, audio, lora_scale, seed, randomize,
     sheet_img = image.convert("RGB").resize((WIDTH, HEIGHT), Image.LANCZOS)
     ref = [sheet_img] * NUM_FRAMES
     pipe.set_adapters("refsheet", float(lora_scale))
-    prompt = _build_prompt(sheet, action, audio)
+    prompt = _build_prompt(sheet, action)
 
     def _cb(p, i, t, kw):
         progress((i + 1) / NUM_STEPS, desc=f"Generating — step {i + 1}/{NUM_STEPS}")
@@ -117,10 +114,8 @@ with gr.Blocks(title="LTX-2.3 Reference Sheet (Fast / Distilled)") as demo:
             image_in = gr.Image(type="pil", label="Reference sheet (one clean panel per element, black background)")
             sheet = gr.Textbox(label="Reference sheet description", lines=3,
                                placeholder="a young woman with red hair in a green jacket (face close-up + turnaround); a brass pocket watch; a cobblestone alley at night")
-            action = gr.Textbox(label="Generated video (the action / shot)", lines=2,
-                                placeholder="the woman walks down the alley and checks the pocket watch, slow dolly-in")
-            audio = gr.Textbox(label="Sound / speech (optional)", lines=1,
-                               placeholder="footsteps on cobblestone, a soft voice saying 'almost time', distant city hum")
+            action = gr.Textbox(label="Generated video — the action / shot, plus any speech & sounds", lines=3,
+                                placeholder="the woman walks down the alley and checks the pocket watch, slow dolly-in; footsteps on cobblestone, a soft voice saying 'almost time', distant city hum")
             with gr.Accordion("Settings", open=False):
                 lora_scale = gr.Slider(0.8, 1.8, value=1.4, step=0.05, label="LoRA strength")
                 randomize = gr.Checkbox(True, label="Randomize seed")
@@ -130,28 +125,25 @@ with gr.Blocks(title="LTX-2.3 Reference Sheet (Fast / Distilled)") as demo:
             video_out = gr.Video(label="Generated video")
             used_seed = gr.Number(label="Seed used", interactive=False)
 
-    run.click(generate, inputs=[image_in, sheet, action, audio, lora_scale, seed, randomize],
+    run.click(generate, inputs=[image_in, sheet, action, lora_scale, seed, randomize],
               outputs=[video_out, used_seed])
 
     gr.Examples(
         examples=[
             ["examples/sheet_camping.jpg",
              "a young child in a yellow raincoat beside a golden retriever dog (large left panel); a cluster of small orange camping tents in a green field (top right); misty green mountains over a calm lake (bottom right)",
-             "the child and the golden retriever walk together across the grassy field toward the tents at dawn, gentle handheld camera, soft morning light",
-             "happy dog panting and a soft bark, a child giggling, birdsong and a gentle breeze",
+             "the child and the golden retriever walk together across the grassy field toward the tents at dawn, gentle handheld camera, soft morning light; happy dog panting and a soft bark, a child giggling, birdsong and a gentle breeze",
              1.4, 42, False],
             ["examples/sheet_astronaut.jpg",
              "an astronaut in a white spacesuit holding the helmet under one arm (large left panel); a vast misty mountain landscape over still water (right panel)",
-             "the astronaut walks slowly across the misty shoreline, looking around in wonder, slow cinematic dolly-in",
-             "crunching footsteps on wet gravel, low wind, soft breathing and a whispered 'incredible'",
+             "the astronaut walks slowly across the misty shoreline, looking around in wonder, slow cinematic dolly-in; crunching footsteps on wet gravel, low wind, soft breathing and a whispered 'incredible'",
              1.4, 42, False],
             ["examples/sheet_woman_horse.jpg",
              "a smiling young woman with curly dark hair in a light top (large left panel); a dappled grey horse standing on grass (top right); a green misty mountain meadow (bottom right)",
-             "the woman walks up to the grey horse in the meadow and gently strokes its neck, smiling, soft daylight",
-             "a woman's warm voice saying 'hey there, good boy', a soft horse nicker, light wind and birdsong",
+             "the woman walks up to the grey horse in the meadow and gently strokes its neck, smiling, soft daylight; a woman's warm voice saying 'hey there, good boy', a soft horse nicker, light wind and birdsong",
              1.4, 42, False],
         ],
-        inputs=[image_in, sheet, action, audio, lora_scale, seed, randomize],
+        inputs=[image_in, sheet, action, lora_scale, seed, randomize],
         outputs=[video_out, used_seed], fn=generate, cache_examples=True, cache_mode="lazy",
     )