Spaces:
Runtime error
Runtime error
File size: 24,631 Bytes
1b92fcf f7c5a08 1b92fcf 9dd5e68 1b92fcf bb7f5b0 1b92fcf |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 |
# Euia-AducSdr: Uma implementação aberta e funcional da arquitetura ADUC-SDR para geração de vídeo coerente.
# Copyright (C) 4 de Agosto de 2025 Carlos Rodrigues dos Santos
#
# Contato:
# Carlos Rodrigues dos Santos
# carlex22@gmail.com
# Rua Eduardo Carlos Pereira, 4125, B1 Ap32, Curitiba, PR, Brazil, CEP 8102025
#
# Repositórios e Projetos Relacionados:
# GitHub: https://github.com/carlex22/Aduc-sdr
# Hugging Face: https://huggingface.co/spaces/Carlexx/Ltx-SuperTime-60Secondos/
# Hugging Face: https://huggingface.co/spaces/Carlexxx/Novinho/
#
# Este programa é software livre: você pode redistribuí-lo e/ou modificá-lo
# sob os termos da Licença Pública Geral Affero da GNU como publicada pela
# Free Software Foundation, seja a versão 3 da Licença, ou
# (a seu critério) qualquer versão posterior.
#
# Este programa é distribuído na esperança de que seja útil,
# mas SEM QUALQUER GARANTIA; sem mesmo a garantia implícita de
# COMERCIALIZAÇÃO ou ADEQUAÇÃO A UM DETERMINADO FIM. Consulte a
# Licença Pública Geral Affero da GNU para mais detalhes.
#
# Você deve ter recebido uma cópia da Licença Pública Geral Affero da GNU
# junto com este programa. Se não, veja <https://www.gnu.org/licenses/>.
# --- app.py (NOVINHO-4.2: Versão Final - Arquitetura "Memória, Caminho, Destino") ---
# --- Ato 1: A Convocação da Orquestra (Importações) ---
import gradio as gr
import torch
import os
import yaml
from PIL import Image, ImageOps
import shutil
import gc
import subprocess
import google.generativeai as genai
import numpy as np
import imageio
from pathlib import Path
import huggingface_hub
import json
import time
from inference import create_ltx_video_pipeline, load_image_to_tensor_with_resize_and_crop, ConditioningItem, calculate_padding
from dreamo_helpers import dreamo_generator_singleton
# --- Ato 2: A Preparação do Palco (Configurações) ---
config_file_path = "configs/ltxv-13b-0.9.8-distilled.yaml"
with open(config_file_path, "r") as file: PIPELINE_CONFIG_YAML = yaml.safe_load(file)
LTX_REPO = "Lightricks/LTX-Video"
models_dir = "downloaded_models_gradio"
Path(models_dir).mkdir(parents=True, exist_ok=True)
WORKSPACE_DIR = "aduc_workspace"
GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY")
VIDEO_FPS = 24
VIDEO_DURATION_SECONDS = 4
VIDEO_TOTAL_FRAMES = VIDEO_DURATION_SECONDS * VIDEO_FPS
CONVERGENCE_FRAMES = 8
TARGET_RESOLUTION = 720
print("Criando pipelines LTX na CPU (estado de repouso)...")
distilled_model_actual_path = huggingface_hub.hf_hub_download(repo_id=LTX_REPO, filename=PIPELINE_CONFIG_YAML["checkpoint_path"], local_dir=models_dir, local_dir_use_symlinks=False)
pipeline_instance = create_ltx_video_pipeline(
ckpt_path=distilled_model_actual_path,
precision=PIPELINE_CONFIG_YAML["precision"],
text_encoder_model_name_or_path=PIPELINE_CONFIG_YAML["text_encoder_model_name_or_path"],
sampler=PIPELINE_CONFIG_YAML["sampler"],
device='cpu'
)
print("Modelos LTX prontos (na CPU).")
# --- Ato 3: As Partituras dos Músicos (Funções Corrigidas, Otimizadas e Documentadas) ---
def load_conditioning_tensor(media_path: str, height: int, width: int) -> torch.Tensor:
if not media_path: raise ValueError("Caminho da mídia de condicionamento não pode ser nulo.")
# A lógica agora só precisa lidar com imagens, simplificando o processo
return load_image_to_tensor_with_resize_and_crop(media_path, height, width)
def run_ltx_animation(current_fragment_index, motion_prompt, conditioning_items_data, width, height, seed, cfg, progress=gr.Progress()):
progress(0, desc=f"[TECPIX 5000] Filmando Cena {current_fragment_index}...");
output_path = os.path.join(WORKSPACE_DIR, f"fragment_{current_fragment_index}_full.mp4");
target_device = pipeline_instance.device
try:
conditioning_items = []
for (path, start_frame, strength) in conditioning_items_data:
tensor = load_conditioning_tensor(path, height, width)
conditioning_items.append(ConditioningItem(tensor.to(target_device), start_frame, strength))
n_val = round((float(VIDEO_TOTAL_FRAMES) - 1.0) / 8.0); actual_num_frames = int(n_val * 8 + 1)
padded_h, padded_w = ((height - 1) // 32 + 1) * 32, ((width - 1) // 32 + 1) * 32
padding_vals = calculate_padding(height, width, padded_h, padded_w)
for cond_item in conditioning_items: cond_item.media_item = torch.nn.functional.pad(cond_item.media_item, padding_vals)
decode_every_val = 4
kwargs = { "prompt": motion_prompt, "negative_prompt": "blurry, distorted, bad quality, artifacts", "height": padded_h, "width": padded_w, "num_frames": actual_num_frames, "frame_rate": VIDEO_FPS, "generator": torch.Generator(device=target_device).manual_seed(int(seed) + current_fragment_index), "output_type": "pt", "guidance_scale": float(cfg), "timesteps": PIPELINE_CONFIG_YAML.get("first_pass", {}).get("timesteps"), "conditioning_items": conditioning_items, "decode_timestep": PIPELINE_CONFIG_YAML.get("decode_timestep"), "decode_noise_scale": PIPELINE_CONFIG_YAML.get("decode_noise_scale"), "stochastic_sampling": PIPELINE_CONFIG_YAML.get("stochastic_sampling"), "image_cond_noise_scale": 0.15, "is_video": True, "vae_per_channel_normalize": True, "mixed_precision": (PIPELINE_CONFIG_YAML.get("precision") == "mixed_precision"), "enhance_prompt": False, "decode_every": decode_every_val }
result_tensor = pipeline_instance(**kwargs).images
pad_l, pad_r, pad_t, pad_b = map(int, padding_vals); slice_h = -pad_b if pad_b > 0 else None; slice_w = -pad_r if pad_r > 0 else None
cropped_tensor = result_tensor[:, :, :VIDEO_TOTAL_FRAMES, pad_t:slice_h, pad_l:slice_w]; video_np = (cropped_tensor[0].permute(1, 2, 3, 0).cpu().float().numpy() * 255).astype(np.uint8)
with imageio.get_writer(output_path, fps=VIDEO_FPS, codec='libx264', quality=8) as writer:
for i, frame in enumerate(video_np):
progress(i / len(video_np), desc=f"Renderizando frame {i+1}/{len(video_np)}...");
writer.append_data(frame)
return output_path, actual_num_frames
except Exception as e:
raise e
def trim_video_to_frames(input_path: str, output_path: str, frames_to_keep: int) -> str:
if not os.path.exists(input_path):
raise gr.Error(f"Erro Interno: Vídeo de entrada para corte não encontrado: {input_path}")
try:
trim_cmd = (f"ffmpeg -y -v error -i \"{input_path}\" -vf \"select='lt(n,{frames_to_keep})'\" -an \"{output_path}\"")
subprocess.run(trim_cmd, shell=True, check=True, capture_output=True, text=True)
return output_path
except subprocess.CalledProcessError as e:
error_message = f"Editor Mágico (FFmpeg) falhou ao cortar o vídeo para {frames_to_keep} frames: {e}"
if hasattr(e, 'stderr'): error_message += f"\nDetalhes: {e.stderr}"
raise gr.Error(error_message)
def extract_last_frame_as_image(video_path: str, output_image_path: str) -> str:
if not os.path.exists(video_path):
raise gr.Error(f"Erro Interno: Vídeo de entrada para extração de frame não encontrado: {video_path}")
try:
command = (f"ffmpeg -y -v error -sseof -1 -i \"{video_path}\" -update 1 -q:v 1 \"{output_image_path}\"")
subprocess.run(command, shell=True, check=True, capture_output=True, text=True)
return output_image_path
except subprocess.CalledProcessError as e:
error_message = f"Editor Mágico (FFmpeg) falhou ao extrair o último frame: {e}"
if hasattr(e, 'stderr'): error_message += f"\nDetalhes: {e.stderr}"
raise gr.Error(error_message)
def process_image_to_square(image_path: str, size: int = TARGET_RESOLUTION) -> str:
if not image_path or not os.path.exists(image_path): return None
try:
img = Image.open(image_path).convert("RGB")
img_square = ImageOps.fit(img, (size, size), Image.Resampling.LANCZOS)
output_filename = f"initial_ref_{size}x{size}.png"
output_path = os.path.join(WORKSPACE_DIR, output_filename)
img_square.save(output_path)
return output_path
except Exception as e: raise gr.Error(f"Falha ao processar a imagem de referência: {e}")
def get_static_scenes_storyboard(num_fragments: int, prompt: str, initial_image_path: str):
if not initial_image_path: raise gr.Error("Por favor, forneça uma imagem de referência inicial.")
if not GEMINI_API_KEY: raise gr.Error("Chave da API Gemini não configurada!")
genai.configure(api_key=GEMINI_API_KEY)
prompt_file = "prompts/photographer_prompt.txt"
with open(os.path.join(os.path.dirname(__file__), prompt_file), "r", encoding="utf-8") as f: template = f.read()
director_prompt = template.format(user_prompt=prompt, num_fragments=int(num_fragments))
model = genai.GenerativeModel('gemini-2.0-flash'); img = Image.open(initial_image_path)
response = model.generate_content([director_prompt, img])
try:
cleaned_response = response.text.strip().replace("```json", "").replace("```", "")
storyboard_data = json.loads(cleaned_response)
return storyboard_data.get("scene_storyboard", [])
except Exception as e: raise gr.Error(f"O Sonhador (Gemini) falhou ao criar o roteiro: {e}. Resposta: {response.text}")
def run_keyframe_generation(storyboard, initial_ref_image_path, sequential_ref_task):
if not storyboard: raise gr.Error("Nenhum roteiro para gerar imagens-chave.")
if not initial_ref_image_path or not os.path.exists(initial_ref_image_path): raise gr.Error("A imagem de referência principal é obrigatória.")
log_history = ""
try:
print("Pintor (DreamO): Movendo a Câmera (LTX) para a CPU para liberar VRAM...")
pipeline_instance.to('cpu')
gc.collect()
if torch.cuda.is_available(): torch.cuda.empty_cache()
print("Pintor (DreamO): VRAM liberada. Movendo o Pintor para a GPU...")
dreamo_generator_singleton.to_gpu()
with Image.open(initial_ref_image_path) as img:
width, height = img.size
width, height = (width // 32) * 32, (height // 32) * 32
keyframe_paths, current_ref_image_path = [], initial_ref_image_path
for i, prompt in enumerate(storyboard):
log_history += f"\nPintando Cena {i+1}/{len(storyboard)}...\n"
yield {keyframe_log_output: gr.update(value=log_history), keyframe_gallery_output: gr.update(value=keyframe_paths)}
reference_items_for_dreamo = [{'image_np': np.array(Image.open(current_ref_image_path).convert("RGB")), 'task': sequential_ref_task}]
log_history += f" - Usando referência: {os.path.basename(current_ref_image_path)} (Tarefa: {sequential_ref_task})\n"
output_path = os.path.join(WORKSPACE_DIR, f"keyframe_{i+1}.png")
image = dreamo_generator_singleton.generate_image_with_gpu_management(reference_items=reference_items_for_dreamo, prompt=prompt, width=width, height=height)
image.save(output_path)
keyframe_paths.append(output_path)
current_ref_image_path = output_path
yield {keyframe_log_output: gr.update(value=log_history), keyframe_gallery_output: gr.update(value=keyframe_paths)}
except Exception as e:
raise gr.Error(f"O Pintor (DreamO) encontrou um erro: {e}")
finally:
print("Pintor (DreamO): Trabalho concluído. Movendo o Pintor de volta para a CPU.")
dreamo_generator_singleton.to_cpu()
gc.collect()
if torch.cuda.is_available(): torch.cuda.empty_cache()
log_history += "\nPintura de todos os keyframes concluída.\n"
yield {keyframe_log_output: gr.update(value=log_history), keyframe_gallery_output: gr.update(value=keyframe_paths), keyframe_images_state: keyframe_paths}
def get_single_motion_prompt(user_prompt: str, story_history: str, start_image_path: str, middle_image_path: str, end_image_path: str):
if not GEMINI_API_KEY: raise gr.Error("Chave da API Gemini não configurada!")
try:
genai.configure(api_key=GEMINI_API_KEY)
model = genai.GenerativeModel('gemini-2.0-flash')
start_img, middle_img, end_img = Image.open(start_image_path), Image.open(middle_image_path), Image.open(end_image_path)
prompt_file_path = os.path.join(os.path.dirname(__file__), "prompts", "director_motion_prompt_three_act.txt")
with open(prompt_file_path, "r", encoding="utf-8") as f:
template = f.read()
director_prompt = template.format(user_prompt=user_prompt, story_history=story_history)
model_contents = [director_prompt, "INÍCIO:", start_img, "MEIO:", middle_img, "FIM:", end_img]
response = model.generate_content(model_contents)
cleaned_text = response.text.strip()
if cleaned_text.startswith("```json"): cleaned_text = cleaned_text[len("```json"):].strip()
if cleaned_text.endswith("```"): cleaned_text = cleaned_text[:-len("```")].strip()
try:
motion_data = json.loads(cleaned_text)
final_prompt = motion_data.get("motion_prompt", "")
if not final_prompt: raise ValueError("Prompt de movimento vazio no JSON.")
return final_prompt
except (json.JSONDecodeError, ValueError):
return cleaned_text.replace("\"", "").replace("{", "").replace("}", "").replace("motion_prompt:", "").strip()
except Exception as e:
response_text = getattr(e, 'text', 'Nenhuma resposta de texto disponível.')
raise gr.Error(f"O Cineasta (Gemini) falhou ao criar o prompt de movimento de 3 atos: {e}. Resposta: {response_text}")
def run_video_production(prompt_geral, keyframe_images_state, scene_storyboard, seed, cfg, cut_frames_value, progress=gr.Progress()):
if not keyframe_images_state or len(keyframe_images_state) < 2:
raise gr.Error("Pinte pelo menos 2 keyframes na Etapa 2 para produzir as transições.")
log_history = "\n--- FASE 3/4: A Câmera e o Cineasta estão filmando em sequência just-in-time...\n"
yield {production_log_output: log_history, video_gallery_glitch: []}
target_device = 'cuda' if torch.cuda.is_available() else 'cpu'
try:
print(f"Câmera (LTX): Movendo para a {target_device} para a produção em lote.")
pipeline_instance.to(target_device)
if target_device == 'cuda':
if hasattr(pipeline_instance, 'disable_model_cpu_offload'): pipeline_instance.disable_model_cpu_offload()
if hasattr(pipeline_instance, 'disable_attention_slicing'): pipeline_instance.disable_attention_slicing()
if hasattr(pipeline_instance.vae, 'disable_slicing'): pipeline_instance.vae.disable_slicing()
if hasattr(pipeline_instance.vae, 'disable_z_tiling'): pipeline_instance.vae.disable_z_tiling()
video_fragments, story_history = [], ""
previous_fragment_last_frame_path = keyframe_images_state[0]
with Image.open(keyframe_images_state[0]) as img: width, height = img.size
num_transitions = len(keyframe_images_state) - 1
for i in range(num_transitions):
start_image_path = previous_fragment_last_frame_path
middle_image_path = keyframe_images_state[i]
end_image_path = keyframe_images_state[i+1]
fragment_num = i + 1
is_last_fragment = (i == num_transitions - 1)
progress(i / num_transitions, desc=f"Planejando e Filmando Fragmento {fragment_num}/{num_transitions}")
log_history += f"\n--- FRAGMENTO {fragment_num} ---\n"
story_history += f"\n- Transição de '{scene_storyboard[i]}' para '{scene_storyboard[i+1]}'."
current_motion_prompt = get_single_motion_prompt(prompt_geral, story_history, start_image_path, middle_image_path, end_image_path)
log_history += f"Instrução do Cineasta (3 Atos): '{current_motion_prompt}'\n"
yield {production_log_output: log_history}
foreshadow_frame, foreshadow_strength = 54, 0.3
end_frame_index = VIDEO_TOTAL_FRAMES - CONVERGENCE_FRAMES
conditioning_items_data = [(start_image_path, 0, 1.0), (end_image_path, foreshadow_frame, foreshadow_strength), (end_image_path, end_frame_index, 1.0)]
full_fragment_path, frames_gerados = run_ltx_animation(fragment_num, current_motion_prompt, conditioning_items_data, width, height, seed, cfg, progress)
log_history += f" - Gerado: {frames_gerados} frames\n"
if not is_last_fragment:
cut_frames = int(cut_frames_value)
final_fragment_path = os.path.join(WORKSPACE_DIR, f"fragment_{fragment_num}_final_{cut_frames}f.mp4")
trim_video_to_frames(full_fragment_path, final_fragment_path, cut_frames)
output_frame_path = os.path.join(WORKSPACE_DIR, f"last_frame_of_frag_{fragment_num}.png")
previous_fragment_last_frame_path = extract_last_frame_as_image(final_fragment_path, output_frame_path)
log_history += f" - Cortado para: {cut_frames} frames\n"
log_history += f" - Memória para próxima cena: Último frame extraído\n"
else:
final_fragment_path = full_fragment_path
log_history += f" - Último fragmento, mantendo duração total: {frames_gerados} frames\n"
video_fragments.append(final_fragment_path)
yield {production_log_output: log_history, video_gallery_glitch: video_fragments}
log_history += "\nFilmagem de todos os fragmentos de transição concluída.\n"
progress(1.0, desc="Produção Concluída.")
yield {production_log_output: log_history, video_gallery_glitch: video_fragments, fragment_list_state: video_fragments}
finally:
print(f"Câmera (LTX): Produção em lote concluída. Movendo para a CPU para liberar VRAM.")
pipeline_instance.to('cpu')
gc.collect()
if torch.cuda.is_available(): torch.cuda.empty_cache()
def concatenate_and_trim_masterpiece(fragment_paths: list, progress=gr.Progress()):
if not fragment_paths: raise gr.Error("Nenhum fragmento de vídeo para concatenar.")
progress(0.5, desc="Montando a obra-prima final...")
try:
list_file_path, final_output_path = os.path.join(WORKSPACE_DIR, "concat_list.txt"), os.path.join(WORKSPACE_DIR, "obra_prima_final.mp4")
with open(list_file_path, "w") as f:
for p in fragment_paths: f.write(f"file '{os.path.abspath(p)}'\n")
concat_cmd = f"ffmpeg -y -v error -f concat -safe 0 -i \"{list_file_path}\" -c copy \"{final_output_path}\""
subprocess.run(concat_cmd, shell=True, check=True, capture_output=True, text=True)
progress(1.0, desc="Montagem concluída!")
return final_output_path
except (subprocess.CalledProcessError, ValueError) as e:
error_message = f"FFmpeg falhou durante a concatenação final: {e}"
if hasattr(e, 'stderr'): error_message += f"\nDetalhes do erro do FFmpeg: {e.stderr}"
raise gr.Error(error_message)
# --- Ato 5: A Interface com o Mundo (A UI Restaurada e Aprimorada) ---
with gr.Blocks(theme=gr.themes.Soft()) as demo:
gr.Markdown("# NOVINHO-4.2 (Piloto de Testes - Arquitetura 'Memória, Caminho, Destino')\n*By Carlex & Gemini & DreamO*")
if os.path.exists(WORKSPACE_DIR): shutil.rmtree(WORKSPACE_DIR)
os.makedirs(WORKSPACE_DIR)
Path("examples").mkdir(exist_ok=True)
scene_storyboard_state, keyframe_images_state, fragment_list_state = gr.State([]), gr.State([]), gr.State([])
prompt_geral_state, processed_ref_path_state = gr.State(""), gr.State("")
gr.Markdown("--- \n ## ETAPA 1: O ROTEIRO (Sonhador)")
with gr.Row():
with gr.Column(scale=1):
prompt_input = gr.Textbox(label="Ideia Geral (Prompt)")
num_fragments_input = gr.Slider(2, 10, 4, step=1, label="Número de Cenas")
image_input = gr.Image(type="filepath", label=f"Imagem de Referência Principal (será {TARGET_RESOLUTION}x{TARGET_RESOLUTION})")
director_button = gr.Button("▶️ 1. Gerar Roteiro de Cenas", variant="primary")
with gr.Column(scale=2): storyboard_to_show = gr.JSON(label="Roteiro de Cenas Gerado")
gr.Markdown("--- \n ## ETAPA 2: OS KEYFRAMES (Pintor)")
with gr.Row():
with gr.Column(scale=2):
gr.Markdown("### Controles do Pintor (DreamO)\n**Tarefas:** `style` (estilo), `ip` (conteúdo), `id` (identidade).")
ref_image_inputs, ref_task_inputs = [], []
with gr.Group():
with gr.Row():
ref_image_inputs.append(gr.Image(label="Referência Inicial / Sequencial (Automática)", type="filepath", interactive=False))
ref_task_inputs.append(gr.Dropdown(choices=["ip", "id", "style"], value="ip", label="Tarefa da Referência"))
photographer_button = gr.Button("▶️ 2. Pintar Imagens-Chave em Cadeia", variant="primary")
with gr.Column(scale=1):
keyframe_log_output = gr.Textbox(label="Diário de Bordo do Pintor", lines=15, interactive=False)
keyframe_gallery_output = gr.Gallery(label="Imagens-Chave Pintadas", object_fit="contain", height="auto", type="filepath")
gr.Markdown("--- \n ## ETAPA 3: A PRODUÇÃO (Cineasta e Câmera)")
with gr.Row():
with gr.Column(scale=1):
with gr.Row():
seed_number = gr.Number(42, label="Seed")
cfg_slider = gr.Slider(1.0, 10.0, 2.5, step=0.1, label="CFG")
cut_frames_slider = gr.Slider(label="Duração do Fragmento (Frames)", minimum=36, maximum=VIDEO_TOTAL_FRAMES, value=90, step=1)
animator_button = gr.Button("▶️ 3. Produzir Cenas em Vídeo", variant="primary")
production_log_output = gr.Textbox(label="Diário de Bordo da Produção", lines=15, interactive=False)
with gr.Column(scale=1): video_gallery_glitch = gr.Gallery(label="Fragmentos Gerados", object_fit="contain", height="auto", type="video")
gr.Markdown(f"--- \n ## ETAPA 4: PÓS-PRODUÇÃO (Editor)")
editor_button = gr.Button("▶️ 4. Montar Vídeo Final", variant="primary")
final_video_output = gr.Video(label="A Obra-Prima Final", width=TARGET_RESOLUTION)
gr.Markdown(
"""
---
### A Arquitetura "Memória, Caminho, Destino"
Nossa geração de vídeo é governada por uma orquestração de IAs, onde cada fragmento (`V_i`) é criado com base em três pilares:
* **Memória (`M_(i-1)`):** O último frame do fragmento anterior. Garante a **continuidade** visual.
* **Caminho (`Γ_i`):** Uma instrução de movimento gerada pelo "Cineasta" (Gemini) ao analisar a Memória, o Keyframe atual e o Destino. Define a **narrativa** da transição.
* **Destino (`K_(i+1)`):** O próximo keyframe a ser alcançado. Define o **objetivo** da animação.
A Câmera (LTX) recebe esses três elementos para construir cada cena, resultando em um vídeo coeso e com propósito.
"""
)
# --- Ato 6: A Regência (Lógica de Conexão dos Botões) ---
director_button.click(
fn=get_static_scenes_storyboard,
inputs=[num_fragments_input, prompt_input, image_input],
outputs=[scene_storyboard_state]
).success(
fn=lambda s, p: (s, p),
inputs=[scene_storyboard_state, prompt_input],
outputs=[storyboard_to_show, prompt_geral_state]
).success(
fn=process_image_to_square,
inputs=[image_input],
outputs=[processed_ref_path_state]
).success(
fn=lambda p: p,
inputs=[processed_ref_path_state],
outputs=[ref_image_inputs[0]]
)
photographer_button.click(
fn=run_keyframe_generation,
inputs=[scene_storyboard_state, processed_ref_path_state, ref_task_inputs[0]],
outputs=[keyframe_log_output, keyframe_gallery_output, keyframe_images_state]
)
animator_button.click(
fn=run_video_production,
inputs=[prompt_geral_state, keyframe_images_state, scene_storyboard_state, seed_number, cfg_slider, cut_frames_slider],
outputs=[production_log_output, video_gallery_glitch, fragment_list_state]
)
editor_button.click(
fn=concatenate_and_trim_masterpiece,
inputs=[fragment_list_state],
outputs=[final_video_output]
)
if __name__ == "__main__":
demo.queue().launch(server_name="0.0.0.0", share=True) |