Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,4 @@
|
|
| 1 |
-
# app.py (VERSÃO FINAL
|
| 2 |
|
| 3 |
import gradio as gr
|
| 4 |
import os
|
|
@@ -7,101 +7,98 @@ import shutil
|
|
| 7 |
import subprocess
|
| 8 |
import mimetypes
|
| 9 |
from pathlib import Path
|
| 10 |
-
from torch.hub import download_url_to_file
|
| 11 |
-
|
| 12 |
-
# --- BLOCO DE CONFIGURAÇÃO E DOWNLOAD DE MODELO CORRIGIDO ---
|
| 13 |
-
APP_DIR = "/app"
|
| 14 |
-
SEEDVR_DIR = os.path.join(APP_DIR, "SeedVR")
|
| 15 |
-
MODEL_CACHE_DIR = "/tmp/models"
|
| 16 |
-
CKPTS_DIR = os.path.join(MODEL_CACHE_DIR, "ckpts")
|
| 17 |
|
|
|
|
|
|
|
|
|
|
| 18 |
os.makedirs(CKPTS_DIR, exist_ok=True)
|
| 19 |
-
|
| 20 |
-
# Dicionário com os links diretos para os arquivos do modelo
|
| 21 |
files_to_download = {
|
| 22 |
"seedvr2_ema_3b.pth": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/seedvr2_ema_3b.pth",
|
| 23 |
"ema_vae.pth": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/ema_vae.pth",
|
| 24 |
"pos_emb.pt": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/pos_emb.pt",
|
| 25 |
"neg_emb.pt": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/neg_emb.pt",
|
| 26 |
}
|
| 27 |
-
|
| 28 |
print("Verificando e baixando modelos para /tmp/models/ckpts...")
|
| 29 |
for filename, url in files_to_download.items():
|
| 30 |
destination_path = os.path.join(CKPTS_DIR, filename)
|
| 31 |
if not os.path.exists(destination_path):
|
| 32 |
-
print(f"Baixando {filename}...")
|
| 33 |
-
|
| 34 |
-
print(f"{filename} baixado com sucesso.")
|
| 35 |
-
else:
|
| 36 |
-
print(f"{filename} já existe. Pulando o download.")
|
| 37 |
print("Verificação de modelos concluída.")
|
| 38 |
# --------------------------------------------------------------------
|
| 39 |
|
| 40 |
def run_inference(video_path, seed, res_h, res_w):
|
| 41 |
-
# ... (O resto do código permanece o mesmo) ...
|
| 42 |
if video_path is None: raise gr.Error("Por favor, faça o upload de um arquivo.")
|
| 43 |
-
|
| 44 |
-
job_id = str(uuid.uuid4())
|
| 45 |
-
input_dir = os.path.join("/tmp", "temp_inputs", job_id)
|
| 46 |
-
output_dir = os.path.join("/tmp", "temp_outputs", job_id)
|
| 47 |
os.makedirs(input_dir, exist_ok=True); os.makedirs(output_dir, exist_ok=True)
|
| 48 |
-
|
| 49 |
shutil.copy(video_path, input_dir)
|
| 50 |
-
|
| 51 |
log_output = ""
|
| 52 |
-
|
| 53 |
patched_script_path = os.path.join("/tmp", f"inference_patched_{job_id}.py")
|
| 54 |
try:
|
| 55 |
original_script_path = os.path.join(SEEDVR_DIR, "projects", "inference_seedvr2_3b.py")
|
| 56 |
with open(original_script_path, 'r') as f: script_content = f.read()
|
| 57 |
-
|
| 58 |
script_content = script_content.replace("'./ckpts/seedvr2_ema_3b.pth'", f"'{os.path.join(CKPTS_DIR, 'seedvr2_ema_3b.pth')}'")
|
| 59 |
script_content = script_content.replace("runner.configure_vae_model()", f"runner.configure_vae_model(checkpoint_path='{os.path.join(CKPTS_DIR, 'ema_vae.pth')}')")
|
| 60 |
script_content = script_content.replace("'pos_emb.pt'", f"'{os.path.join(CKPTS_DIR, 'pos_emb.pt')}'")
|
| 61 |
script_content = script_content.replace("'neg_emb.pt'", f"'{os.path.join(CKPTS_DIR, 'neg_emb.pt')}'")
|
| 62 |
-
|
| 63 |
with open(patched_script_path, 'w') as f: f.write(script_content)
|
| 64 |
-
|
| 65 |
-
input_folder_relative = os.path.relpath(input_dir, SEEDVR_DIR)
|
| 66 |
-
output_folder_relative = os.path.relpath(output_dir, SEEDVR_DIR)
|
| 67 |
patched_script_relative_path = os.path.relpath(patched_script_path, SEEDVR_DIR)
|
| 68 |
-
|
| 69 |
command = ["torchrun", "--nproc-per-node=4", patched_script_relative_path, "--video_path", input_folder_relative, "--output_dir", output_folder_relative, "--seed", str(seed), "--res_h", str(res_h), "--res_w", str(res_w)]
|
| 70 |
-
|
| 71 |
env = os.environ.copy(); env["PYTHONUNBUFFERED"] = "1"
|
| 72 |
-
|
| 73 |
log_output += f"Executando comando: {' '.join(command)}\n\n"
|
| 74 |
yield None, None, log_output
|
| 75 |
-
|
| 76 |
process = subprocess.Popen(command, cwd=SEEDVR_DIR, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True, encoding='utf-8', env=env)
|
| 77 |
-
|
| 78 |
while True:
|
| 79 |
output = process.stdout.readline()
|
| 80 |
if output == '' and process.poll() is not None: break
|
| 81 |
-
if output:
|
| 82 |
-
log_output += output
|
| 83 |
-
yield None, None, log_output
|
| 84 |
-
|
| 85 |
if process.poll() != 0: raise gr.Error(f"A inferência falhou.")
|
| 86 |
-
|
| 87 |
output_files = [f for f in os.listdir(output_dir) if f.endswith(('.mp4', '.png', '.jpg', '.jpeg'))]
|
| 88 |
if not output_files: raise gr.Error("Nenhum arquivo de saída foi encontrado.")
|
| 89 |
-
|
| 90 |
result_path = os.path.join(output_dir, output_files[0])
|
| 91 |
-
|
| 92 |
media_type, _ = mimetypes.guess_type(result_path)
|
| 93 |
if media_type and media_type.startswith("image"): yield result_path, None, log_output
|
| 94 |
else: yield None, result_path, log_output
|
| 95 |
-
|
| 96 |
finally:
|
| 97 |
shutil.rmtree(input_dir, ignore_errors=True)
|
| 98 |
if os.path.exists(patched_script_path): os.remove(patched_script_path)
|
| 99 |
|
| 100 |
-
# --- Interface Gráfica Gradio
|
| 101 |
with gr.Blocks(css="footer {display: none !important}") as demo:
|
| 102 |
gr.Markdown("# 🚀 Interface de Inferência para SeedVR2")
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
gr.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
|
| 107 |
demo.queue(max_size=10).launch()
|
|
|
|
| 1 |
+
# app.py (VERSÃO FINAL E CORRIGIDA)
|
| 2 |
|
| 3 |
import gradio as gr
|
| 4 |
import os
|
|
|
|
| 7 |
import subprocess
|
| 8 |
import mimetypes
|
| 9 |
from pathlib import Path
|
| 10 |
+
from torch.hub import download_url_to_file
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
+
# --- BLOCO DE CONFIGURAÇÃO E DOWNLOAD DE MODELO ---
|
| 13 |
+
APP_DIR = "/app"; SEEDVR_DIR = os.path.join(APP_DIR, "SeedVR")
|
| 14 |
+
MODEL_CACHE_DIR = "/tmp/models"; CKPTS_DIR = os.path.join(MODEL_CACHE_DIR, "ckpts")
|
| 15 |
os.makedirs(CKPTS_DIR, exist_ok=True)
|
|
|
|
|
|
|
| 16 |
files_to_download = {
|
| 17 |
"seedvr2_ema_3b.pth": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/seedvr2_ema_3b.pth",
|
| 18 |
"ema_vae.pth": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/ema_vae.pth",
|
| 19 |
"pos_emb.pt": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/pos_emb.pt",
|
| 20 |
"neg_emb.pt": "https://huggingface.co/ByteDance-Seed/SeedVR2-3B/resolve/main/neg_emb.pt",
|
| 21 |
}
|
|
|
|
| 22 |
print("Verificando e baixando modelos para /tmp/models/ckpts...")
|
| 23 |
for filename, url in files_to_download.items():
|
| 24 |
destination_path = os.path.join(CKPTS_DIR, filename)
|
| 25 |
if not os.path.exists(destination_path):
|
| 26 |
+
print(f"Baixando {filename}..."); download_url_to_file(url, destination_path)
|
| 27 |
+
else: print(f"{filename} já existe.")
|
|
|
|
|
|
|
|
|
|
| 28 |
print("Verificação de modelos concluída.")
|
| 29 |
# --------------------------------------------------------------------
|
| 30 |
|
| 31 |
def run_inference(video_path, seed, res_h, res_w):
|
|
|
|
| 32 |
if video_path is None: raise gr.Error("Por favor, faça o upload de um arquivo.")
|
| 33 |
+
job_id = str(uuid.uuid4()); input_dir = os.path.join("/tmp", "temp_inputs", job_id); output_dir = os.path.join("/tmp", "temp_outputs", job_id)
|
|
|
|
|
|
|
|
|
|
| 34 |
os.makedirs(input_dir, exist_ok=True); os.makedirs(output_dir, exist_ok=True)
|
|
|
|
| 35 |
shutil.copy(video_path, input_dir)
|
|
|
|
| 36 |
log_output = ""
|
|
|
|
| 37 |
patched_script_path = os.path.join("/tmp", f"inference_patched_{job_id}.py")
|
| 38 |
try:
|
| 39 |
original_script_path = os.path.join(SEEDVR_DIR, "projects", "inference_seedvr2_3b.py")
|
| 40 |
with open(original_script_path, 'r') as f: script_content = f.read()
|
|
|
|
| 41 |
script_content = script_content.replace("'./ckpts/seedvr2_ema_3b.pth'", f"'{os.path.join(CKPTS_DIR, 'seedvr2_ema_3b.pth')}'")
|
| 42 |
script_content = script_content.replace("runner.configure_vae_model()", f"runner.configure_vae_model(checkpoint_path='{os.path.join(CKPTS_DIR, 'ema_vae.pth')}')")
|
| 43 |
script_content = script_content.replace("'pos_emb.pt'", f"'{os.path.join(CKPTS_DIR, 'pos_emb.pt')}'")
|
| 44 |
script_content = script_content.replace("'neg_emb.pt'", f"'{os.path.join(CKPTS_DIR, 'neg_emb.pt')}'")
|
|
|
|
| 45 |
with open(patched_script_path, 'w') as f: f.write(script_content)
|
| 46 |
+
input_folder_relative = os.path.relpath(input_dir, SEEDVR_DIR); output_folder_relative = os.path.relpath(output_dir, SEEDVR_DIR)
|
|
|
|
|
|
|
| 47 |
patched_script_relative_path = os.path.relpath(patched_script_path, SEEDVR_DIR)
|
|
|
|
| 48 |
command = ["torchrun", "--nproc-per-node=4", patched_script_relative_path, "--video_path", input_folder_relative, "--output_dir", output_folder_relative, "--seed", str(seed), "--res_h", str(res_h), "--res_w", str(res_w)]
|
|
|
|
| 49 |
env = os.environ.copy(); env["PYTHONUNBUFFERED"] = "1"
|
|
|
|
| 50 |
log_output += f"Executando comando: {' '.join(command)}\n\n"
|
| 51 |
yield None, None, log_output
|
|
|
|
| 52 |
process = subprocess.Popen(command, cwd=SEEDVR_DIR, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True, encoding='utf-8', env=env)
|
|
|
|
| 53 |
while True:
|
| 54 |
output = process.stdout.readline()
|
| 55 |
if output == '' and process.poll() is not None: break
|
| 56 |
+
if output: log_output += output; yield None, None, log_output
|
|
|
|
|
|
|
|
|
|
| 57 |
if process.poll() != 0: raise gr.Error(f"A inferência falhou.")
|
|
|
|
| 58 |
output_files = [f for f in os.listdir(output_dir) if f.endswith(('.mp4', '.png', '.jpg', '.jpeg'))]
|
| 59 |
if not output_files: raise gr.Error("Nenhum arquivo de saída foi encontrado.")
|
|
|
|
| 60 |
result_path = os.path.join(output_dir, output_files[0])
|
|
|
|
| 61 |
media_type, _ = mimetypes.guess_type(result_path)
|
| 62 |
if media_type and media_type.startswith("image"): yield result_path, None, log_output
|
| 63 |
else: yield None, result_path, log_output
|
|
|
|
| 64 |
finally:
|
| 65 |
shutil.rmtree(input_dir, ignore_errors=True)
|
| 66 |
if os.path.exists(patched_script_path): os.remove(patched_script_path)
|
| 67 |
|
| 68 |
+
# --- Interface Gráfica Gradio ---
|
| 69 |
with gr.Blocks(css="footer {display: none !important}") as demo:
|
| 70 |
gr.Markdown("# 🚀 Interface de Inferência para SeedVR2")
|
| 71 |
+
gr.Markdown("Faça o upload de um vídeo ou imagem, ajuste os parâmetros e clique em 'Executar'.")
|
| 72 |
+
|
| 73 |
+
with gr.Row():
|
| 74 |
+
with gr.Column(scale=1):
|
| 75 |
+
input_media = gr.Video(label="Upload de Vídeo ou Imagem")
|
| 76 |
+
seed = gr.Number(value=666, label="Seed")
|
| 77 |
+
with gr.Accordion("Configurações Avançadas", open=False):
|
| 78 |
+
res_h = gr.Number(value=720, label="Altura da Saída (res_h)")
|
| 79 |
+
res_w = gr.Number(value=1280, label="Largura da Saída (res_w)")
|
| 80 |
+
run_button = gr.Button("Executar", variant="primary")
|
| 81 |
+
|
| 82 |
+
with gr.Column(scale=2):
|
| 83 |
+
output_image = gr.Image(label="Saída de Imagem")
|
| 84 |
+
output_video = gr.Video(label="Saída de Vídeo")
|
| 85 |
+
log_box = gr.Textbox(label="Logs em Tempo Real", lines=15, autoscroll=True, interactive=False)
|
| 86 |
+
|
| 87 |
+
# !!! A CORREÇÃO FINAL ESTÁ AQUI !!!
|
| 88 |
+
# Estas duas chamadas foram movidas para DENTRO do bloco 'with gr.Blocks()'.
|
| 89 |
+
run_button.click(
|
| 90 |
+
fn=run_inference,
|
| 91 |
+
inputs=[input_media, seed, res_h, res_w],
|
| 92 |
+
outputs=[output_image, output_video, log_box]
|
| 93 |
+
)
|
| 94 |
+
|
| 95 |
+
gr.Examples(
|
| 96 |
+
examples=[
|
| 97 |
+
["./SeedVR/01.mp4", 666],
|
| 98 |
+
["./SeedVR/02.mp4", 123],
|
| 99 |
+
["./SeedVR/03.mp4", 42],
|
| 100 |
+
],
|
| 101 |
+
inputs=[input_media, seed]
|
| 102 |
+
)
|
| 103 |
|
| 104 |
demo.queue(max_size=10).launch()
|