Upload app_seedvr.py

app_seedvr.py

@@ -0,0 +1,117 @@
+#!/usr/bin/env python3
+"""
+SeedVR UI (Gradio) — CLI torchrun
+
+- Upload único: vídeo (.mp4) ou imagem (.png/.jpg/.jpeg/.webp).
+- Parâmetros: seed, res_h, res_w, sp_size.
+- Executa via torchrun com NUM_GPUS (do ambiente).
+- Exibe vídeo se a entrada for vídeo; imagem se for imagem.
+"""
+
+import os
+import mimetypes
+from pathlib import Path
+from typing import Optional
+
+import gradio as gr
+
+from services.seed_server import SeedVRServer
+
+# Instância única do servidor (clona repo, baixa modelo, cria symlink)
+server = SeedVRServer()
+
+# Paths padrão (para allowed_paths e debug)
+OUTPUT_ROOT = Path(os.getenv("OUTPUT_ROOT", "/app/outputs"))
+CKPTS_ROOT = Path(os.getenv("CKPTS_ROOT", "/app/ckpts/SeedVR2-3B"))
+
+def _is_video(path: str) -> bool:
+    mime, _ = mimetypes.guess_type(path)
+    return (mime or "").startswith("video") or str(path).lower().endswith(".mp4")
+
+def _is_image(path: str) -> bool:
+    mime, _ = mimetypes.guess_type(path)
+    if mime and mime.startswith("image"):
+        return True
+    return str(path).lower().endswith((".png", ".jpg", ".jpeg", ".webp"))
+
+def ui_infer(
+    input_path: Optional[str],
+    seed: int,
+    res_h: int,
+    res_w: int,
+    sp_size: int,
+):
+    if not input_path or not Path(input_path).exists():
+        gr.Warning("Arquivo de entrada ausente ou inválido.")
+        return None, None
+
+    is_vid = _is_video(input_path)
+    is_img = _is_image(input_path)
+    if not (is_vid or is_img):
+        gr.Warning("Tipo de arquivo não suportado. Envie .mp4, .png, .jpg, .jpeg ou .webp.")
+        return None, None
+
+    try:
+        video_out, image_out, _ = server.run_inference(
+            file_path=input_path,
+            seed=int(seed),
+            res_h=int(res_h),
+            res_w=int(res_w),
+            sp_size=int(sp_size),
+        )
+    except Exception as e:
+        gr.Warning(f"Erro na inferência: {e}")
+        return None, None
+
+    if is_vid:
+        if video_out and Path(video_out).exists():
+            return None, video_out
+        if image_out and Path(image_out).exists():
+            return image_out, None
+        gr.Warning("Nenhum resultado encontrado.")
+        return None, None
+    else:
+        if image_out and Path(image_out).exists():
+            return image_out, None
+        if video_out and Path(video_out).exists():
+            return None, video_out
+        gr.Warning("Nenhum resultado encontrado.")
+        return None, None
+
+with gr.Blocks(title="SeedVR (CLI torchrun)") as demo:
+    gr.Markdown(
+        "\n".join([
+            "# SeedVR — Restauração (CLI torchrun)",
+            "- Envie um vídeo (.mp4) ou uma imagem (.png/.jpg/.jpeg/.webp).",
+            "- A execução utiliza torchrun com múltiplas GPUs.",
+        ])
+    )
+
+    with gr.Row():
+        inp = gr.File(label="Entrada (vídeo .mp4 ou imagem)", type="filepath")
+
+    with gr.Row():
+        seed = gr.Number(label="Seed", value=int(os.getenv("SEED", "42")), precision=0)
+        res_h = gr.Number(label="Altura (res_h)", value=int(os.getenv("RES_H", "720")), precision=0)
+        res_w = gr.Number(label="Largura (res_w)", value=int(os.getenv("RES_W", "1280")), precision=0)
+        sp_size = gr.Number(label="sp_size", value=int(os.getenv("SP_SIZE", "4")), precision=0)
+
+    run = gr.Button("Restaurar", variant="primary")
+
+    out_image = gr.Image(label="Resultado (imagem)")
+    out_video = gr.Video(label="Resultado (vídeo)")
+
+    run.click(
+        ui_infer,
+        inputs=[inp, seed, res_h, res_w, sp_size],
+        outputs=[out_image, out_video],
+    )
+
+if __name__ == "__main__":
+    demo.launch(
+        server_name=os.getenv("GRADIO_SERVER_NAME", "0.0.0.0"),
+        server_port=int(os.getenv("GRADIO_SERVER_PORT", os.getenv("PORT", "7860"))),
+        allowed_paths=[str(OUTPUT_ROOT), str(CKPTS_ROOT)],
+        show_error=True,
+    )
+