Update app_vince.py

app_vince.py

@@ -2,57 +2,47 @@
 """
 VINCIE Service UI (Gradio)
 
-- Automatic setup runs on app load (no manual setup button).
-- Multi-turn editing and multi-concept composition front-end.
-- Designed for NVIDIA L40S (SM 8.9) environments aligned with CUDA 12.x.
-- Functional reference: ByteDance-Seed/VINCIE.
-- Space and Docker developed by Carlex (contact below).
+- Auto-setup na carga da interface (sem botão manual).
+- Edição multi-turn e composição multi-concept com overrides avançados (steps, cfg_scale, aspect_ratio, resolution).
+- Repassa parâmetros para o pipeline do VINCIE através do VincieService.
+- Desenvolvido para ambientes NVIDIA L40S (SM 8.9) em CUDA 12.x.
+
+Upstream funcional: ByteDance-Seed/VINCIE.
 """
 
 import os
+import json
 from pathlib import Path
 from typing import List, Tuple, Optional
 
 import gradio as gr
 
-# Adapt this import to the project layout.
-# Provide a VincieService with:
-# - ensure_repo(): clones/updates upstream repo if missing
-# - ensure_model(): downloads/validates checkpoints to /app/ckpt/VINCIE-3B
-# - multi_turn_edit(image_path: str, turns: List[str], *, cfg_scale: Optional[float], resolution_input: Optional[int], aspect_ratio_input: Optional[str], steps: Optional[int]) -> str
-# - multi_concept_compose(files: List[str], descs: List[str], final_prompt: str, *, cfg_scale: Optional[float], resolution_input: Optional[int], aspect_ratio_input: Optional[str], steps: Optional[int]) -> str
-from services.vincie import VincieService  # change path if needed
+# Serviço que prepara repositório e checkpoint, e invoca geração por CLI
+from services.vincie import VincieService  # ajuste o import conforme seu layout
 
-# Instantiate the service (defaults to /app/VINCIE and /app/ckpt/VINCIE-3B)
+# Instancia o serviço (defaults: /app/VINCIE repo e cache HF/HUB já configurado externamente)
 svc = VincieService()
 
+
 def setup_auto() -> str:
     """
-    Run an idempotent setup on interface load:
-    - Ensure the upstream repository is present and ready.
-    - Ensure the model checkpoint is downloaded and ready.
-    Returns an English status string for the UI.
+    Setup idempotente na carga da UI:
+    - Clona/Checa o repositório upstream.
+    - Baixa/Valida o checkpoint (snapshot completo) no cache HF.
     """
     try:
         svc.ensure_repo()
         svc.ensure_model()
-        return (
-            "Setup completed successfully: repository and checkpoint are ready "
-            "for inference on an NVIDIA L40S environment."
-        )
+        if not svc.ready():
+            return "Setup finalizado, mas checkpoint incompleto (dit.pth/vae.pth ausentes)."  # pragma: no cover
+        return "Setup concluído: repositório e checkpoint prontos para inferência."  # ok
     except Exception as e:
-        return f"Setup encountered an error: {e}"
+        return f"Erro no setup: {e}"
+
 
 def _list_media(out_dir: Path, max_images: int = 24) -> Tuple[List[str], Optional[str]]:
     """
-    Enumerate resulting images and the most recent video from an output directory.
-    Args:
-        out_dir: Path to the directory where the service wrote its results.
-        max_images: Upper bound on how many images to surface in the gallery.
-    Returns:
-        A tuple (images, video) where:
-        - images is a list of file paths to images sorted by modified time,
-        - video is the path to the latest .mp4 if found, otherwise None.
+    Coleta imagens e o último .mp4 do diretório de saída.
     """
     img_globs = ("*.png", "*.jpg", "*.jpeg", "*.webp")
     images: List[Path] = []
@@ -64,6 +54,7 @@ def _list_media(out_dir: Path, max_images: int = 24) -> Tuple[List[str], Optiona
     video_path = str(videos[-1]) if videos else None
     return image_paths, video_path
 
+
 def ui_multi_turn(
     input_image: Optional[str],
     turns_text: Optional[str],
@@ -73,21 +64,12 @@ def ui_multi_turn(
     resolution_input: int,
 ):
     """
-    Multi-turn image editing entrypoint for the UI.
-    Args:
-        input_image: Path to a single input image on disk.
-        turns_text: User-provided editing turns, one instruction per line.
-        steps_input: Override for generation steps.
-        cfg_scale_input: Override for guidance scale.
-        aspect_ratio_input: Override for aspect ratio selection.
-        resolution_input: Override for shorter-side resolution.
-    Returns:
-        (gallery, video, status) for Gradio components.
+    Handler de UI para multi-turn editing.
     """
     if not input_image or not str(input_image).strip():
-        return [], None, "Please provide an input image."
+        return [], None, "Selecione uma imagem de entrada."
     if not turns_text or not turns_text.strip():
-        return [], None, "Please provide edit turns (one per line)."
+        return [], None, "Informe instruções (uma por linha)."
 
     turns = [ln.strip() for ln in turns_text.splitlines() if ln.strip()]
     try:
@@ -100,12 +82,13 @@ def ui_multi_turn(
             steps=int(steps_input) if steps_input is not None else None,
         )
     except Exception as e:
-        return [], None, f"Generation error: {e}"
+        return [], None, f"Erro na geração: {e}"
 
     imgs, vid = _list_media(Path(out_dir))
-    status = f"Outputs saved to: {out_dir}"
+    status = f"Outputs salvos em: {out_dir}"
     return imgs, vid, status
 
+
 def ui_multi_concept(
     files: Optional[List[str]],
     descs_text: Optional[str],
@@ -116,28 +99,18 @@ def ui_multi_concept(
     resolution_input: int,
 ):
     """
-    Multi-concept composition entrypoint for the UI.
-    Args:
-        files: List of paths to concept images on disk.
-        descs_text: Per-image descriptions (one line per image, in the same order).
-        final_prompt: A final composition prompt that aggregates the concepts.
-        steps_input: Override for generation steps.
-        cfg_scale_input: Override for guidance scale.
-        aspect_ratio_input: Override for aspect ratio selection.
-        resolution_input: Override for shorter-side resolution.
-    Returns:
-        (gallery, video, status) for Gradio components.
+    Handler de UI para multi-concept composition.
     """
     if not files:
-        return [], None, "Please upload concept images."
+        return [], None, "Envie imagens de conceito."
     if not descs_text or not descs_text.strip():
-        return [], None, "Please provide descriptions (one per line)."
+        return [], None, "Descreva os conceitos (uma linha por imagem)."
     if not final_prompt or not final_prompt.strip():
-        return [], None, "Please provide a final prompt."
+        return [], None, "Informe um prompt final."
 
     descs = [ln.strip() for ln in descs_text.splitlines() if ln.strip()]
     if len(descs) != len(files):
-        return [], None, f"Descriptions count ({len(descs)}) must match images count ({len(files)})."
+        return [], None, f"Número de descrições ({len(descs)}) deve bater com imagens ({len(files)})."
 
     try:
         out_dir = svc.multi_concept_compose(
@@ -150,31 +123,30 @@ def ui_multi_concept(
             steps=int(steps_input) if steps_input is not None else None,
         )
     except Exception as e:
-        return [], None, f"Generation error: {e}"
+        return [], None, f"Erro na geração: {e}"
 
     imgs, vid = _list_media(Path(out_dir))
-    status = f"Outputs saved to: {out_dir}"
+    status = f"Outputs salvos em: {out_dir}"
     return imgs, vid, status
 
+
 with gr.Blocks(title="VINCIE Service") as demo:
-    # Header and credits
+    # Header
     gr.Markdown(
         "\n".join(
             [
-                "# VINCIE Service — Multi-turn Editing and Multi-concept Composition",
-                "- Automatic setup runs at startup; setup status appears below.",
-                "- Hardware requirement: NVIDIA L40S (SM 8.9) is recommended for this build.",
-                "- Functional upstream model: ByteDance-Seed/VINCIE (see project repository).",
-                "- Space and Docker were developed by Carlex.",
-                "- Contact: Email: [email protected] | GitHub: carlex22",
+                "# VINCIE — Edição Multi-Turn e Composição Multi-Concept",
+                "- Setup automático é executado no carregamento; verifique o status abaixo.",
+                "- Requisitos: NVIDIA L40S (SM 8.9) recomendado para este build.",
+                "- Upstream: ByteDance-Seed/VINCIE.",
             ]
         )
     )
 
     with gr.Row():
-        setup_out = gr.Textbox(label="Setup Status", interactive=False)
+        setup_out = gr.Textbox(label="Status do Setup", interactive=False)
 
-    # Advanced options shared by both tabs
+    # Opções avançadas compartilhadas entre abas
     with gr.Accordion("Opções Avançadas (para Abas 1 e 2)", open=False):
         steps_input = gr.Slider(
             label="Passos de Inferência",
@@ -203,14 +175,15 @@ with gr.Blocks(title="VINCIE Service") as demo:
             value=512,
         )
 
+    # Aba 1 — Multi-turn Editing
     with gr.Tab("Multi-turn Editing"):
         with gr.Row():
-            img = gr.Image(type="filepath", label="Input image")
-            turns = gr.Textbox(lines=8, label="Turns (one per line)")
-        run1 = gr.Button("Run")
-        out_gallery = gr.Gallery(label="Images", columns=4, height="auto")
-        out_video = gr.Video(label="Video (if available)")
-        out_status = gr.Textbox(label="Output", interactive=False)
+            img = gr.Image(type="filepath", label="Imagem de entrada")
+            turns = gr.Textbox(lines=8, label="Instruções (uma por linha)")
+        run1 = gr.Button("Gerar")
+        out_gallery = gr.Gallery(label="Imagens", columns=4, height="auto")
+        out_video = gr.Video(label="Vídeo (se houver)")
+        out_status = gr.Textbox(label="Saída", interactive=False)
 
         run1.click(
             ui_multi_turn,
@@ -218,14 +191,15 @@ with gr.Blocks(title="VINCIE Service") as demo:
             outputs=[out_gallery, out_video, out_status],
         )
 
+    # Aba 2 — Multi-concept Composition
     with gr.Tab("Multi-concept Composition"):
-        files = gr.File(file_count="multiple", type="filepath", label="Concept images")
-        descs = gr.Textbox(lines=8, label="Descriptions (one per line, same order as images)")
-        final_prompt = gr.Textbox(lines=2, label="Final prompt")
-        run2 = gr.Button("Run")
-        out_gallery2 = gr.Gallery(label="Images", columns=4, height="auto")
-        out_video2 = gr.Video(label="Video (if available)")
-        out_status2 = gr.Textbox(label="Output", interactive=False)
+        files = gr.File(file_count="multiple", type="filepath", label="Imagens de conceito")
+        descs = gr.Textbox(lines=8, label="Descrições (uma por linha, mesma ordem das imagens)")
+        final_prompt = gr.Textbox(lines=2, label="Prompt final")
+        run2 = gr.Button("Gerar")
+        out_gallery2 = gr.Gallery(label="Imagens", columns=4, height="auto")
+        out_video2 = gr.Video(label="Vídeo (se houver)")
+        out_status2 = gr.Textbox(label="Saída", interactive=False)
 
         run2.click(
             ui_multi_concept,
@@ -233,13 +207,13 @@ with gr.Blocks(title="VINCIE Service") as demo:
             outputs=[out_gallery2, out_video2, out_status2],
         )
 
-    # Auto-setup on load (no manual button)
+    # Auto-setup on load
     demo.load(fn=setup_auto, outputs=setup_out)
 
 if __name__ == "__main__":
     demo.launch(
-        server_name="0.0.0.0",
-        server_port=int(os.getenv("PORT", "7860")),
+        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=["/app/outputs", "/app/ckpt"],
+        enable_queue=os.getenv("GRADIO_ENABLE_QUEUE", "True").lower() == "true",
     )
-