Update app.py

app.py

@@ -480,79 +480,23 @@ def _replacement_command(
 
 
 @spaces.GPU(duration=GPU_DURATION, size=GPU_SIZE)
-def build_animation_pack(
+def build_simple_pack(
+    mode,
     ref_image,
     driving_video,
     prompt,
     sam3_text,
     max_persons,
     crop_mode,
-    progress=gr.Progress(track_tqdm=True),
-):
-    try:
-        progress(0.0, desc="Checking SCAIL-Pose repo")
-        _require_repo_layout()
-        progress(0.04, desc="Preparing SAM3 weights")
-        sam3_model = _ensure_sam3_weights()
-
-        job_dir = _new_job_dir("animation")
-        progress(0.08, desc="Preparing inputs")
-        _save_reference_image(ref_image, job_dir / "ref.png")
-        _copy_video(driving_video, job_dir / "driving.mp4", "driving video")
-        _write_metadata(
-            job_dir,
-            "animation",
-            prompt,
-            {
-                "driving": {
-                    "video": "rendered_v2.mp4",
-                    "mask_video": "rendered_mask_v2.mp4",
-                },
-                "primary": {
-                    "image": "ref.png",
-                    "mask": "ref_mask.jpg",
-                },
-                "sam3_text": _text_args(sam3_text),
-                "max_persons": int(max_persons),
-                "crop_mode": crop_mode,
-            },
-        )
-
-        progress(0.12, desc="Generating animation masks")
-        logs = _run_command(
-            _animation_command(job_dir, sam3_model, int(max_persons), sam3_text, crop_mode),
-            progress=progress,
-        )
-        _require_outputs(job_dir, ["ref_mask.jpg", "rendered_v2.mp4", "rendered_mask_v2.mp4"])
-
-        progress(0.92, desc="Packaging SCAIL-2 input pack")
-        zip_path = _zip_pack(job_dir, "animation")
-        progress(1.0, desc="Done")
-        status = f"Done. Pack created at {zip_path}\n\n{logs}"
-        return (
-            str(job_dir / "ref_mask.jpg"),
-            str(job_dir / "rendered_v2.mp4"),
-            str(job_dir / "rendered_mask_v2.mp4"),
-            str(zip_path),
-            status,
-        )
-    except Exception:
-        logging.exception("Animation pack generation failed")
-        return None, None, None, None, traceback.format_exc()
-
-
-@spaces.GPU(duration=GPU_DURATION, size=GPU_SIZE)
-def build_replacement_pack(
-    ref_image,
-    driving_video,
-    prompt,
-    sam3_text,
     matchnearest,
     egocentric,
     progress=gr.Progress(track_tqdm=True),
 ):
     try:
-        if matchnearest and egocentric:
+        mode = str(mode or "Animation").lower()
+        if mode not in {"animation", "replacement"}:
+            raise RuntimeError(f"Unsupported mode: {mode}")
+        if mode == "replacement" and matchnearest and egocentric:
             raise RuntimeError("matchnearest and egocentric are mutually exclusive.")
 
         progress(0.0, desc="Checking SCAIL-Pose repo")
@@ -560,8 +504,50 @@ def build_replacement_pack(
         progress(0.04, desc="Preparing SAM3 weights")
         sam3_model = _ensure_sam3_weights()
 
+        if mode == "animation":
+            job_dir = _new_job_dir("animation")
+            progress(0.08, desc="Preparing animation inputs")
+            _save_reference_image(ref_image, job_dir / "ref.png")
+            _copy_video(driving_video, job_dir / "driving.mp4", "driving video")
+            _write_metadata(
+                job_dir,
+                "animation",
+                prompt,
+                {
+                    "driving": {
+                        "video": "rendered_v2.mp4",
+                        "mask_video": "rendered_mask_v2.mp4",
+                    },
+                    "primary": {
+                        "image": "ref.png",
+                        "mask": "ref_mask.jpg",
+                    },
+                    "sam3_text": _text_args(sam3_text),
+                    "max_persons": int(max_persons),
+                    "crop_mode": crop_mode,
+                },
+            )
+
+            progress(0.12, desc="Generating animation masks")
+            logs = _run_command(
+                _animation_command(job_dir, sam3_model, int(max_persons), sam3_text, crop_mode),
+                progress=progress,
+            )
+            _require_outputs(job_dir, ["ref_mask.jpg", "rendered_v2.mp4", "rendered_mask_v2.mp4"])
+            progress(0.92, desc="Packaging SCAIL-2 input pack")
+            zip_path = _zip_pack(job_dir, "animation")
+            progress(1.0, desc="Done")
+            status = f"Done. Animation pack created at {zip_path}\n\n{logs}"
+            return (
+                str(job_dir / "ref_mask.jpg"),
+                str(job_dir / "rendered_v2.mp4"),
+                str(job_dir / "rendered_mask_v2.mp4"),
+                str(zip_path),
+                status,
+            )
+
         job_dir = _new_job_dir("replacement")
-        progress(0.08, desc="Preparing inputs")
+        progress(0.08, desc="Preparing replacement inputs")
         _save_reference_image(ref_image, job_dir / "ref.png")
         _copy_video(driving_video, job_dir / "driving.mp4", "source video")
         _write_metadata(
@@ -589,11 +575,10 @@ def build_replacement_pack(
             progress=progress,
         )
         _require_outputs(job_dir, ["ref_mask.png", "rendered_v2.mp4", "replace_mask.mp4"])
-
         progress(0.92, desc="Packaging SCAIL-2 input pack")
         zip_path = _zip_pack(job_dir, "replacement")
         progress(1.0, desc="Done")
-        status = f"Done. Pack created at {zip_path}\n\n{logs}"
+        status = f"Done. Replacement pack created at {zip_path}\n\n{logs}"
         return (
             str(job_dir / "ref_mask.png"),
             str(job_dir / "rendered_v2.mp4"),
@@ -602,7 +587,7 @@ def build_replacement_pack(
             status,
         )
     except Exception:
-        logging.exception("Replacement pack generation failed")
+        logging.exception("Simple pack generation failed")
         return None, None, None, None, traceback.format_exc()
 
 
@@ -742,42 +727,59 @@ def build_ui():
         )
         gr.Textbox(value=_repo_status(), label="Startup status", interactive=False, lines=7)
 
-        with gr.Tab("Animation Pack"):
+        with gr.Tab("Simple Pack"):
             gr.Markdown(
-                "Create an animation pack from one reference image and one driving video. "
-                "This uses SCAIL-Pose end-to-end mode: the driving video becomes `rendered_v2.mp4`, "
-                "and SAM3 produces the colored driving mask video."
+                "Create a standard SCAIL-2 pack from one reference image and one video. "
+                "Choose Animation to follow motion from the driving video, or Replacement to replace "
+                "a target region in the source video."
+            )
+            simple_mode = gr.Radio(
+                ["Animation", "Replacement"],
+                value="Animation",
+                label="Mode",
             )
             with gr.Row():
-                anim_ref = gr.Image(type="filepath", label="Reference image")
-                anim_driving = gr.Video(label="Driving video")
-            anim_prompt = gr.Textbox(label="Prompt for SCAIL-2", lines=3)
-            with gr.Row():
-                anim_sam3_text = gr.Textbox(value="human character", label="SAM3 text prompt")
-                anim_max_persons = gr.Number(value=2, precision=0, label="Max tracked subjects")
-                anim_crop = gr.Dropdown(
+                simple_ref = gr.Image(type="filepath", label="Reference image")
+                simple_driving = gr.Video(label="Driving / source video")
+            simple_prompt = gr.Textbox(label="Prompt for SCAIL-2", lines=3)
+            simple_sam3_text = gr.Textbox(value="human character", label="SAM3 text prompt")
+            with gr.Accordion("Animation options", open=True):
+                simple_max_persons = gr.Number(value=2, precision=0, label="Max tracked subjects")
+                simple_crop = gr.Dropdown(
                     ["none", "mask silhouette", "moving bbox", "steady bbox"],
                     value="none",
                     label="Driving crop mode",
                 )
-            anim_run = gr.Button("Generate animation pack", variant="primary")
+            with gr.Accordion("Replacement options", open=False):
+                simple_matchnearest = gr.Checkbox(
+                    value=False,
+                    label="Match nearest target when two people are detected",
+                )
+                simple_egocentric = gr.Checkbox(
+                    value=False,
+                    label="Egocentric: union two detected parts into one actor",
+                )
+            simple_run = gr.Button("Generate pack", variant="primary")
             with gr.Row():
-                anim_ref_mask = gr.Image(label="Reference mask", interactive=False)
-                anim_mask_video = gr.Video(label="Driving mask video")
-            anim_rendered = gr.Video(label="Rendered / driving video")
-            anim_zip = gr.File(label="Download SCAIL-2 animation pack")
-            anim_status = gr.Textbox(label="Run logs", lines=14)
-            anim_run.click(
-                build_animation_pack,
+                simple_ref_mask = gr.Image(label="Reference mask", interactive=False)
+                simple_mask_video = gr.Video(label="Mask video")
+            simple_rendered = gr.Video(label="Rendered / source video")
+            simple_zip = gr.File(label="Download SCAIL-2 pack")
+            simple_status = gr.Textbox(label="Run logs", lines=14)
+            simple_run.click(
+                build_simple_pack,
                 inputs=[
-                    anim_ref,
-                    anim_driving,
-                    anim_prompt,
-                    anim_sam3_text,
-                    anim_max_persons,
-                    anim_crop,
+                    simple_mode,
+                    simple_ref,
+                    simple_driving,
+                    simple_prompt,
+                    simple_sam3_text,
+                    simple_max_persons,
+                    simple_crop,
+                    simple_matchnearest,
+                    simple_egocentric,
                 ],
-                outputs=[anim_ref_mask, anim_rendered, anim_mask_video, anim_zip, anim_status],
+                outputs=[simple_ref_mask, simple_rendered, simple_mask_video, simple_zip, simple_status],
             )
 
         with gr.Tab("Advanced Character Pack"):
@@ -848,42 +850,9 @@ def build_ui():
                 outputs=[adv_gallery, adv_rendered, adv_mask_video, adv_zip, adv_status],
             )
 
-        with gr.Tab("Replacement Pack"):
-            gr.Markdown(
-                "Create a replacement pack from a replacement reference image and a source video. "
-                "The source video becomes `rendered_v2.mp4`, and SAM3 produces `replace_mask.mp4`."
-            )
-            with gr.Row():
-                repl_ref = gr.Image(type="filepath", label="Replacement reference image")
-                repl_driving = gr.Video(label="Source / driving video")
-            repl_prompt = gr.Textbox(label="Prompt for SCAIL-2", lines=3)
-            with gr.Row():
-                repl_sam3_text = gr.Textbox(value="human character", label="SAM3 text prompt")
-                repl_matchnearest = gr.Checkbox(value=False, label="Match nearest target when two people are detected")
-                repl_egocentric = gr.Checkbox(value=False, label="Egocentric: union two detected parts into one actor")
-            repl_run = gr.Button("Generate replacement pack", variant="primary")
-            with gr.Row():
-                repl_ref_mask = gr.Image(label="Reference mask", interactive=False)
-                repl_mask_video = gr.Video(label="Replacement mask video")
-            repl_rendered = gr.Video(label="Rendered / source video")
-            repl_zip = gr.File(label="Download SCAIL-2 replacement pack")
-            repl_status = gr.Textbox(label="Run logs", lines=14)
-            repl_run.click(
-                build_replacement_pack,
-                inputs=[
-                    repl_ref,
-                    repl_driving,
-                    repl_prompt,
-                    repl_sam3_text,
-                    repl_matchnearest,
-                    repl_egocentric,
-                ],
-                outputs=[repl_ref_mask, repl_rendered, repl_mask_video, repl_zip, repl_status],
-            )
-
         with gr.Tab("Pack Format"):
             gr.Markdown(
-                "The simple tabs export flat packs. The advanced tab exports a canonical "
+                "The Simple Pack tab exports flat packs. The Advanced Character Pack tab exports a canonical "
                 "character pack with explicit character folders.\n\n"
                 "Simple animation pack:\n"
                 "```text\n"