# app.py (Versão Corrigida) import gradio as gr from PIL import Image import os import imageio from api.ltx_server import video_generation_service from huggingface_hub import logging logging.set_verbosity_error() logging.set_verbosity_warning() logging.set_verbosity_info() logging.set_verbosity_debug() # --- FUNÇÕES DE AJUDA PARA A UI --- # ... (calculate_new_dimensions e handle_media_upload_for_dims permanecem as mesmas) ... TARGET_FIXED_SIDE = 768 MIN_DIM_SLIDER = 256 MAX_IMAGE_SIZE = 1280 def calculate_new_dimensions(orig_w, orig_h): if orig_w == 0 or orig_h == 0: return int(TARGET_FIXED_SIDE), int(TARGET_FIXED_SIDE) if orig_w >= orig_h: new_h, aspect_ratio = TARGET_FIXED_SIDE, orig_w / orig_h new_w = round((new_h * aspect_ratio) / 32) * 32 new_w = max(MIN_DIM_SLIDER, min(new_w, MAX_IMAGE_SIZE)) new_h = max(MIN_DIM_SLIDER, min(new_h, MAX_IMAGE_SIZE)) else: new_w, aspect_ratio = TARGET_FIXED_SIDE, orig_h / orig_w new_h = round((new_w * aspect_ratio) / 32) * 32 new_h = max(MIN_DIM_SLIDER, min(new_h, MAX_IMAGE_SIZE)) new_w = max(MIN_DIM_SLIDER, min(new_w, MAX_IMAGE_SIZE)) return int(new_h), int(new_w) def handle_media_upload_for_dims(filepath, current_h, current_w): if not filepath or not os.path.exists(str(filepath)): return gr.update(value=current_h), gr.update(value=current_w) try: if str(filepath).lower().endswith(('.png', '.jpg', '.jpeg', '.webp')): with Image.open(filepath) as img: orig_w, orig_h = img.size else: # Assumir que é um vídeo with imageio.get_reader(filepath) as reader: meta = reader.get_meta_data() orig_w, orig_h = meta.get('size', (current_w, current_h)) new_h, new_w = calculate_new_dimensions(orig_w, orig_h) return gr.update(value=new_h), gr.update(value=new_w) except Exception as e: print(f"Erro ao processar mídia para dimensões: {e}") return gr.update(value=current_h), gr.update(value=current_w) def update_frame_slider(duration): """Atualiza o valor máximo do slider de frame do meio com base na duração.""" fps = 24.0 max_frames = int(duration * fps) # Garante que o valor padrão não seja maior que o novo máximo new_value = 48 if max_frames >= 48 else max_frames // 2 return gr.update(maximum=max_frames, value=new_value) # --- FUNÇÃO WRAPPER PARA CHAMAR O SERVIÇO --- def gradio_generate_wrapper( prompt, negative_prompt, mode, # Entradas de Keyframe start_image, middle_image, middle_frame, middle_weight, end_image, end_weight, # Outras entradas input_video, height, width, duration, frames_to_use, seed, randomize_seed, guidance_scale, improve_texture, progress=gr.Progress(track_tqdm=True) ): try: def progress_handler(step, total_steps): progress(step / total_steps, desc="Salvando vídeo...") output_path, used_seed = video_generation_service.generate( prompt=prompt, negative_prompt=negative_prompt, mode=mode, start_image_filepath=start_image, middle_image_filepath=middle_image, middle_frame_number=middle_frame, middle_image_weight=middle_weight, end_image_filepath=end_image, end_image_weight=end_weight, input_video_filepath=input_video, height=int(height), width=int(width), duration=float(duration), frames_to_use=int(frames_to_use), seed=int(seed), randomize_seed=bool(randomize_seed), guidance_scale=float(guidance_scale), improve_texture=bool(improve_texture), progress_callback=progress_handler ) return output_path, used_seed except ValueError as e: raise gr.Error(str(e)) except Exception as e: print(f"Erro inesperado na geração: {e}") raise gr.Error("Ocorreu um erro inesperado. Verifique os logs.") # --- DEFINIÇÃO DA INTERFACE GRADIO --- css = "#col-container { margin: 0 auto; max-width: 900px; }" with gr.Blocks(css=css) as demo: gr.Markdown("# LTX Video com Keyframes") gr.Markdown("Guie a geração de vídeo usando imagens de início, meio e fim.") with gr.Row(): with gr.Column(): with gr.Tab("image-to-video (Keyframes)") as image_tab: i2v_prompt = gr.Textbox(label="Prompt", value="Uma bela transição entre as imagens", lines=2) with gr.Row(): with gr.Column(scale=1): gr.Markdown("#### Início (Obrigatório)") start_image_i2v = gr.Image(label="Imagem de Início", type="filepath", sources=["upload", "clipboard"]) with gr.Row(): with gr.Accordion("Imagens condicionais opcional", open=False): with gr.Column(scale=1): gr.Markdown("#### Meio (Opcional)") middle_image_i2v = gr.Image(label="Imagem do Meio", type="filepath", sources=["upload", "clipboard"]) middle_frame_i2v = gr.Slider(label="Frame Alvo", minimum=0, maximum=200, step=1, value=48) middle_weight_i2v = gr.Slider(label="Peso/Força", minimum=0.0, maximum=1.0, step=0.05, value=1.0) with gr.Column(scale=1): gr.Markdown("#### Fim (Opcional)") end_image_i2v = gr.Image(label="Imagem de Fim", type="filepath", sources=["upload", "clipboard"]) end_weight_i2v = gr.Slider(label="Peso/Força", minimum=0.0, maximum=1.0, step=0.05, value=1.0) i2v_button = gr.Button("Generate Image-to-Video", variant="primary") with gr.Tab("text-to-video") as text_tab: t2v_prompt = gr.Textbox(label="Prompt", value="A majestic dragon flying over a medieval castle", lines=3) t2v_button = gr.Button("Generate Text-to-Video", variant="primary") with gr.Tab("video-to-video") as video_tab: video_v2v = gr.Video(label="Input Video", sources=["upload", "webcam"]) frames_to_use = gr.Slider(label="Frames to use from input video", minimum=9, maximum=257, value=9, step=8, info="Must be N*8+1.") v2v_prompt = gr.Textbox(label="Prompt", value="Change the style to cinematic anime", lines=3) v2v_button = gr.Button("Generate Video-to-Video", variant="primary") duration_input = gr.Slider(label="Video Duration (seconds)", minimum=1, maximum=30, value=8, step=0.5) improve_texture = gr.Checkbox(label="Improve Texture (multi-scale)", value=True, visible=True) with gr.Column(): output_video = gr.Video(label="Generated Video", interactive=False) with gr.Accordion("Advanced settings", open=False): mode = gr.Dropdown(["text-to-video", "image-to-video", "video-to-video"], label="task", value="image-to-video", visible=False) negative_prompt_input = gr.Textbox(label="Negative Prompt", value="worst quality, blurry, jittery", lines=2) with gr.Row(): seed_input = gr.Number(label="Seed", value=42, precision=0) randomize_seed_input = gr.Checkbox(label="Randomize Seed", value=True) guidance_scale_input = gr.Slider(label="Guidance Scale (CFG)", minimum=1.0, maximum=10.0, value=3.0, step=0.1) with gr.Row(): height_input = gr.Slider(label="Height", value=512, step=32, minimum=MIN_DIM_SLIDER, maximum=MAX_IMAGE_SIZE) width_input = gr.Slider(label="Width", value=704, step=32, minimum=MIN_DIM_SLIDER, maximum=MAX_IMAGE_SIZE) # --- LÓGICA DE EVENTOS DA UI --- start_image_i2v.upload(fn=handle_media_upload_for_dims, inputs=[start_image_i2v, height_input, width_input], outputs=[height_input, width_input]) video_v2v.upload(fn=handle_media_upload_for_dims, inputs=[video_v2v, height_input, width_input], outputs=[height_input, width_input]) duration_input.change(fn=update_frame_slider, inputs=duration_input, outputs=middle_frame_i2v) image_tab.select(fn=lambda: "image-to-video", outputs=[mode]) text_tab.select(fn=lambda: "text-to-video", outputs=[mode]) video_tab.select(fn=lambda: "video-to-video", outputs=[mode]) # Placeholders para os botões que não usam certos inputs none_image = gr.Textbox(visible=False, value=None) none_video = gr.Textbox(visible=False, value=None) # Parâmetros comuns a todos shared_params = [ height_input, width_input, duration_input, frames_to_use, seed_input, randomize_seed_input, guidance_scale_input, improve_texture ] i2v_inputs = [ i2v_prompt, negative_prompt_input, mode, start_image_i2v, middle_image_i2v, middle_frame_i2v, middle_weight_i2v, end_image_i2v, end_weight_i2v, none_video, # Placeholder para input_video *shared_params ] t2v_inputs = [ t2v_prompt, negative_prompt_input, mode, none_image, none_image, gr.Number(value=-1, visible=False), gr.Slider(value=0, visible=False), # Placeholders para keyframes none_image, gr.Slider(value=0, visible=False), none_video, # Placeholder para input_video *shared_params ] v2v_inputs = [ v2v_prompt, negative_prompt_input, mode, none_image, none_image, gr.Number(value=-1, visible=False), gr.Slider(value=0, visible=False), # Placeholders para keyframes none_image, gr.Slider(value=0, visible=False), video_v2v, # Input de vídeo real *shared_params ] common_outputs = [output_video, seed_input] i2v_button.click(fn=gradio_generate_wrapper, inputs=i2v_inputs, outputs=common_outputs, api_name="image_to_video_keyframes") t2v_button.click(fn=gradio_generate_wrapper, inputs=t2v_inputs, outputs=common_outputs, api_name="text_to_video") v2v_button.click(fn=gradio_generate_wrapper, inputs=v2v_inputs, outputs=common_outputs, api_name="video_to_video") # --- --- if __name__ == "__main__": demo.queue().launch(debug=True, share=False)