import gradio as gr import numpy as np import random import torch import spaces from PIL import Image from diffusers import FlowMatchEulerDiscreteScheduler from optimization import optimize_pipeline_ from qwenimage.pipeline_qwenimage_edit_plus import QwenImageEditPlusPipeline from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel from qwenimage.qwen_fa3_processor import QwenDoubleStreamAttnProcessorFA3 import math from huggingface_hub import hf_hub_download from safetensors.torch import load_file from PIL import Image import os import gradio as gr from gradio_client import Client, handle_file import tempfile from huggingface_hub import InferenceClient # --- Model Loading --- dtype = torch.bfloat16 device = "cuda" if torch.cuda.is_available() else "cpu" scheduler_config = { "base_image_seq_len": 256, "base_shift": math.log(3), "invert_sigmas": False, "max_image_seq_len": 8192, "max_shift": math.log(3), "num_train_timesteps": 1000, "shift": 1.0, "shift_terminal": None, "stochastic_sampling": False, "time_shift_type": "exponential", "use_beta_sigmas": False, "use_dynamic_shifting": True, "use_exponential_sigmas": False, "use_karras_sigmas": False, } scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config) pipe = QwenImageEditPlusPipeline.from_pretrained("Qwen/Qwen-Image-Edit-2509", scheduler=scheduler, torch_dtype=dtype) # Load the relight LoRA pipe.load_lora_weights("dx8152/Qwen-Image-Edit-2509-Relight", weight_name="Qwen-Edit-Relight.safetensors", adapter_name="relight") pipe.load_lora_weights("lightx2v/Qwen-Image-Lightning", weight_name="Qwen-Image-Lightning-4steps-V2.0-bf16.safetensors", adapter_name="lightning") pipe.set_adapters(["relight", "lightning"], adapter_weights=[1., 1.]) pipe.fuse_lora(adapter_names=["relight", "lightning"], lora_scale=1) pipe.unload_lora_weights() pipe.transformer.__class__ = QwenImageTransformer2DModel pipe.transformer.set_attn_processor(QwenDoubleStreamAttnProcessorFA3()) pipe.to(device) optimize_pipeline_(pipe, image=[Image.new("RGB", (1024, 1024)), Image.new("RGB", (1024, 1024))], prompt="prompt") MAX_SEED = np.iinfo(np.int32).max translation_client = InferenceClient( api_key=os.environ.get("HF_TOKEN"), ) def translate_to_chinese(text: str) -> str: """Translate any language text to Chinese using Qwen API.""" if not text or not text.strip(): return "" # Check if text is already primarily Chinese chinese_chars = sum(1 for char in text if '\u4e00' <= char <= '\u9fff') if chinese_chars / max(len(text), 1) > 0.5: # Already mostly Chinese, return as is return text try: completion = translation_client.chat.completions.create( model="Qwen/Qwen3-Next-80B-A3B-Instruct:novita", messages=[ { "role": "system", "content": "You are a professional translator. Translate the user's text to Chinese. Only output the translated text, nothing else." }, { "role": "user", "content": f"Translate this to Chinese: {text}" } ], max_tokens=500, ) translated = completion.choices[0].message.content.strip() print(f"Translated '{text}' to '{translated}'") return translated except Exception as e: print(f"Translation error: {e}") # Fallback to original text if translation fails return text def _generate_video_segment(input_image_path: str, output_image_path: str, prompt: str, request: gr.Request) -> str: """Generates a single video segment using the external service.""" x_ip_token = request.headers['x-ip-token'] video_client = Client("multimodalart/wan-2-2-first-last-frame", headers={"x-ip-token": x_ip_token}) result = video_client.predict( start_image_pil=handle_file(input_image_path), end_image_pil=handle_file(output_image_path), prompt=prompt, api_name="/generate_video", ) return result[0]["video"] def build_relight_prompt(light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt): """Build the relighting prompt based on user selections - Qwen style.""" # Priority 1: User's prompt (translated to Chinese if needed) if prompt and prompt.strip(): translated = translate_to_chinese(prompt) # Add trigger word if not already present if "重新照明" not in translated: return f"重新照明,{translated}" return translated # Priority 2: Build from controls prompt_parts = ["重新照明"] # Light type descriptions (expanded from IC-Light style but in Chinese) light_descriptions = { "none": "", "soft_window": "窗帘透光（柔和漫射）", "golden_hour": "金色黄昏的温暖光线", "studio": "专业摄影棚的均匀光线", "dramatic": "戏剧性的高对比度光线", "natural": "自然日光", "neon": "霓虹灯光效果", "candlelight": "烛光的温暖氛围", "moonlight": "月光的冷色调", "sunrise": "日出的柔和光线", "sunset_sea": "海面日落光线", "overcast": "阴天的柔和漫射光", "harsh_sun": "强烈的正午阳光", "twilight": "黄昏时分的神秘光线", "aurora": "极光般的多彩光线", "firelight": "篝火的跳动光线", "lightning": "闪电的瞬间强光", "underwater": "水下的柔和蓝光", "foggy": "雾气中的柔和扩散光", "magic": "魔法般的神秘光芒", "cyberpunk": "赛博朋克风格的RGB霓虹光", "warm_home": "家庭温馨的暖色光", "cold_industrial": "冷酷的工业照明", "spotlight": "聚光灯效果", "rim_light": "边缘光效果", } # Direction descriptions (from IC-Light) direction_descriptions = { "none": "", "front": "正面照射", "side": "侧面照射", "left": "左侧照射", "right": "右侧照射", "back": "背后照射（逆光）", "top": "上方照射", "bottom": "下方照射", "diagonal": "对角线照射", } # Intensity descriptions intensity_descriptions = { "none": "", "soft": "柔和强度", "medium": "中等强度", "strong": "强烈强度", } # Illumination environments (from IC-Light vary, translated) illumination_envs = { "none": "", "sunshine_window": "阳光从窗户透入", "neon_city": "霓虹夜景，城市灯光", "sci_fi_rgb": "科幻RGB发光，赛博朋克风格", "warm_bedroom": "温暖氛围，家中，卧室", "magic_lit": "魔法照明", "gothic_cave": "邪恶哥特风格，洞穴中", "light_shadow": "光影交错", "window_shadow": "窗户投影", "soft_studio": "柔和摄影棚灯光", "cozy_bedroom": "家庭氛围，温馨卧室照明", "wong_kar_wai": "王家卫风格霓虹灯，温暖色调", "moonlight_curtains": "月光透过窗帘", "stormy_sky": "暴风雨天空照明", "underwater_glow": "水下发光，深海", "foggy_forest": "雾中森林黎明", "meadow_golden": "草地上的黄金时刻", "rainbow_neon": "彩虹反射，霓虹", "apocalyptic": "末日烟雾氛围", "emergency_red": "红色紧急灯光", "mystical_forest": "神秘发光，魔法森林", "campfire": "篝火光芒", "industrial_harsh": "严酷工业照明", "mountain_sunrise": "山中日出", "desert_evening": "沙漠黄昏", "dark_alley": "黑暗小巷的月光", "fairground": "游乐场的金色光芒", "forest_midnight": "森林深夜", "twilight_purple": "黄昏的紫粉色调", "foggy_morning": "雾蒙蒙的早晨", "rustic_candle": "乡村风格烛光", "office_fluorescent": "办公室荧光灯", "storm_lightning": "暴风雨中的闪电", "fireplace_night": "夜晚壁炉的温暖光芒", "ethereal_magic": "空灵发光，魔法森林", "beach_dusky": "海滩的黄昏", "trees_afternoon": "树林中的午后光线", "urban_blue_neon": "蓝色霓虹灯，城市街道", "rain_police": "雨中红蓝警灯", "aurora_arctic": "极光，北极景观", "foggy_mountains": "雾中山峦日出", "city_skyline": "城市天际线的黄金时刻", "twilight_mist": "神秘黄昏，浓雾", "forest_rays": "森林空地的清晨光线", "festival_lantern": "节日多彩灯笼光", "stained_glass": "彩色玻璃的柔和光芒", "dark_spotlight": "黑暗房间的强烈聚光", "lake_evening": "湖面柔和的黄昏光", "cave_crystal": "洞穴水晶反射", "autumn_forest": "秋林中的鲜艳光线", "snowfall_dusk": "黄昏轻柔降雪", "winter_hazy": "冬日清晨的朦胧光", "rain_city": "雨中城市灯光倒影", "trees_golden_sun": "金色阳光穿过树林", "fireflies_summer": "萤火虫点亮夏夜", "forge_embers": "锻造炉的发光余烬", "gothic_castle": "哥特城堡的昏暗烛光", "starlight_midnight": "午夜明亮星光", "rural_sunset": "乡村的温暖日落", "haunted_flicker": "闹鬼房屋的闪烁灯光", "desert_mirage": "沙漠日落海市蜃楼般的光", "storm_beams": "风暴云中穿透的金色光束", } # Build the prompt - Qwen style (comma-separated, Chinese) # Handle custom light type if light_type == "custom" and light_type_custom and light_type_custom.strip(): prompt_parts.append(translate_to_chinese(light_type_custom)) elif light_type != "none": prompt_parts.append(light_descriptions.get(light_type, "")) # Handle custom illumination environment if illumination_env == "custom" and illumination_env_custom and illumination_env_custom.strip(): prompt_parts.append(translate_to_chinese(illumination_env_custom)) elif illumination_env != "none": prompt_parts.append(illumination_envs.get(illumination_env, "")) # Handle custom light direction if light_direction == "custom" and light_direction_custom and light_direction_custom.strip(): prompt_parts.append(translate_to_chinese(light_direction_custom)) elif light_direction != "none": prompt_parts.append(direction_descriptions.get(light_direction, "")) # Handle custom light intensity if light_intensity == "custom" and light_intensity_custom and light_intensity_custom.strip(): prompt_parts.append(translate_to_chinese(light_intensity_custom)) elif light_intensity != "none": prompt_parts.append(intensity_descriptions.get(light_intensity, "")) final_prompt = ",".join([p for p in prompt_parts if p]) # Add instruction if we have settings if len(prompt_parts) > 1: final_prompt += ",对图片进行重新照明" return final_prompt if len(prompt_parts) > 1 else "重新照明,使用自然光线对图片进行重新照明" @spaces.GPU def infer_relight( image, light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt, seed, randomize_seed, true_guidance_scale, num_inference_steps, height, width, prev_output = None, progress=gr.Progress(track_tqdm=True) ): final_prompt = build_relight_prompt(light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt) print(f"Generated Prompt: {final_prompt}") if randomize_seed: seed = random.randint(0, MAX_SEED) generator = torch.Generator(device=device).manual_seed(seed) # Choose input image (prefer uploaded, else last output) pil_images = [] if image is not None: if isinstance(image, Image.Image): pil_images.append(image.convert("RGB")) elif hasattr(image, "name"): pil_images.append(Image.open(image.name).convert("RGB")) elif prev_output: pil_images.append(prev_output.convert("RGB")) if len(pil_images) == 0: raise gr.Error("Please upload an image first.") result = pipe( image=pil_images, prompt=final_prompt, height=height if height != 0 else None, width=width if width != 0 else None, num_inference_steps=num_inference_steps, generator=generator, true_cfg_scale=true_guidance_scale, num_images_per_prompt=1, ).images[0] return result, seed, final_prompt def create_video_between_images(input_image, output_image, prompt: str, request: gr.Request) -> str: """Create a video between the input and output images.""" if input_image is None or output_image is None: raise gr.Error("Both input and output images are required to create a video.") try: with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp: input_image.save(tmp.name) input_image_path = tmp.name output_pil = Image.fromarray(output_image.astype('uint8')) with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp: output_pil.save(tmp.name) output_image_path = tmp.name video_path = _generate_video_segment( input_image_path, output_image_path, prompt if prompt else "Relighting transformation", request ) return video_path except Exception as e: raise gr.Error(f"Video generation failed: {e}") # --- UI --- css = ''' #col-container { max-width: 1200px; margin: 0 auto; } .dark .progress-text{color: white !important} #examples{max-width: 1200px; margin: 0 auto; } .radio-group {display: grid; grid-template-columns: repeat(auto-fit, minmax(200px, 1fr)); gap: 8px;} .radio-group [data-testid="block-info"] { display: none !important } ''' def reset_all(): return ["none", "", "none", "", "none", "", "none", "", "", False] def end_reset(): return False def update_dimensions_on_upload(image): if image is None: return 1024, 1024 original_width, original_height = image.size if original_width > original_height: new_width = 1024 aspect_ratio = original_height / original_width new_height = int(new_width * aspect_ratio) else: new_height = 1024 aspect_ratio = original_width / original_height new_width = int(new_height * aspect_ratio) # Ensure dimensions are multiples of 8 new_width = (new_width // 8) * 8 new_height = (new_height // 8) * 8 return new_width, new_height def toggle_custom_textbox(choice): """Show textbox when Custom is selected""" return gr.update(visible=(choice == "custom")) with gr.Blocks(theme=gr.themes.Citrus(), css=css) as demo: with gr.Column(elem_id="col-container"): gr.Markdown("## 💡 Qwen Image Edit — Relighting Control") gr.Markdown(""" Qwen Image Edit 2509 for Image Relighting ✨ Using [dx8152's Qwen-Image-Edit-2509-Relight LoRA](https://huggingface.co/dx8152/Qwen-Image-Edit-2509-Relight) and [lightx2v/Qwen-Image-Lightning](https://huggingface.co/lightx2v/Qwen-Image-Lightning) for 4-step inference 💨 """ ) with gr.Row(): with gr.Column(scale=1): image = gr.Image(label="Input Image", type="pil") prev_output = gr.Image(value=None, visible=False) is_reset = gr.Checkbox(value=False, visible=False) with gr.Tab("Compose Prompt"): with gr.Accordion("💡 Light Type", open=True): light_type = gr.Radio( choices=[ ("None", "none"), ("Soft Window Light", "soft_window"), ("Golden Hour", "golden_hour"), ("Studio Lighting", "studio"), ("Dramatic", "dramatic"), ("Natural Daylight", "natural"), ("Neon", "neon"), ("Candlelight", "candlelight"), ("Moonlight", "moonlight"), ("Sunrise", "sunrise"), ("Sunset over Sea", "sunset_sea"), ("Overcast", "overcast"), ("Harsh Sunlight", "harsh_sun"), ("Twilight", "twilight"), ("Aurora", "aurora"), ("Firelight", "firelight"), ("Lightning", "lightning"), ("Underwater", "underwater"), ("Foggy", "foggy"), ("Magic Light", "magic"), ("Cyberpunk", "cyberpunk"), ("Warm Home", "warm_home"), ("Cold Industrial", "cold_industrial"), ("Spotlight", "spotlight"), ("Rim Light", "rim_light"), ("Custom", "custom"), ], value="none", elem_classes="radio-group" ) light_type_custom = gr.Textbox( label="Custom Light Type", placeholder="e.g., Bioluminescent glow, Laser light show, etc.", visible=False ) with gr.Accordion("🧭 Light Direction", open=True): light_direction = gr.Radio( choices=[ ("None", "none"), ("Front", "front"), ("Side", "side"), ("Left", "left"), ("Right", "right"), ("Back (Backlight)", "back"), ("Top", "top"), ("Bottom", "bottom"), ("Diagonal", "diagonal"), ("Custom", "custom"), ], value="none", elem_classes="radio-group" ) light_direction_custom = gr.Textbox( label="Custom Light Direction", placeholder="e.g., From 45 degrees above left, Rotating around subject, etc.", visible=False ) with gr.Accordion("⚡ Light Intensity", open=True): light_intensity = gr.Radio( choices=[ ("None", "none"), ("Soft", "soft"), ("Medium", "medium"), ("Strong", "strong"), ("Custom", "custom"), ], value="none", elem_classes="radio-group" ) light_intensity_custom = gr.Textbox( label="Custom Light Intensity", placeholder="e.g., Very dim, Blinding bright, Pulsating, etc.", visible=False ) with gr.Accordion("🌍 Illumination Environment", open=False): illumination_env = gr.Radio( choices=[ ("None", "none"), ("Sunshine from Window", "sunshine_window"), ("Neon Night, City", "neon_city"), ("Sci-Fi RGB Glowing, Cyberpunk", "sci_fi_rgb"), ("Warm Atmosphere, at Home, Bedroom", "warm_bedroom"), ("Magic Lit", "magic_lit"), ("Evil, Gothic, in a Cave", "gothic_cave"), ("Light and Shadow", "light_shadow"), ("Shadow from Window", "window_shadow"), ("Soft Studio Lighting", "soft_studio"), ("Home Atmosphere, Cozy Bedroom", "cozy_bedroom"), ("Neon, Wong Kar-wai, Warm", "wong_kar_wai"), ("Moonlight through Curtains", "moonlight_curtains"), ("Stormy Sky Lighting", "stormy_sky"), ("Underwater Glow, Deep Sea", "underwater_glow"), ("Foggy Forest at Dawn", "foggy_forest"), ("Golden Hour in a Meadow", "meadow_golden"), ("Rainbow Reflections, Neon", "rainbow_neon"), ("Apocalyptic, Smoky Atmosphere", "apocalyptic"), ("Red Glow, Emergency Lights", "emergency_red"), ("Mystical Glow, Enchanted Forest", "mystical_forest"), ("Campfire Light", "campfire"), ("Harsh, Industrial Lighting", "industrial_harsh"), ("Sunrise in the Mountains", "mountain_sunrise"), ("Evening Glow in the Desert", "desert_evening"), ("Moonlight in a Dark Alley", "dark_alley"), ("Golden Glow at a Fairground", "fairground"), ("Midnight in the Forest", "forest_midnight"), ("Purple and Pink Hues at Twilight", "twilight_purple"), ("Foggy Morning, Muted Light", "foggy_morning"), ("Candle-lit Room, Rustic Vibe", "rustic_candle"), ("Fluorescent Office Lighting", "office_fluorescent"), ("Lightning Flash in Storm", "storm_lightning"), ("Night, Cozy Warm Light from Fireplace", "fireplace_night"), ("Ethereal Glow, Magical Forest", "ethereal_magic"), ("Dusky Evening on a Beach", "beach_dusky"), ("Afternoon Light Filtering through Trees", "trees_afternoon"), ("Blue Neon Light, Urban Street", "urban_blue_neon"), ("Red and Blue Police Lights in Rain", "rain_police"), ("Aurora Borealis Glow, Arctic Landscape", "aurora_arctic"), ("Sunrise through Foggy Mountains", "foggy_mountains"), ("Golden Hour on a City Skyline", "city_skyline"), ("Mysterious Twilight, Heavy Mist", "twilight_mist"), ("Early Morning Rays, Forest Clearing", "forest_rays"), ("Colorful Lantern Light at Festival", "festival_lantern"), ("Soft Glow through Stained Glass", "stained_glass"), ("Harsh Spotlight in Dark Room", "dark_spotlight"), ("Mellow Evening Glow on a Lake", "lake_evening"), ("Crystal Reflections in a Cave", "cave_crystal"), ("Vibrant Autumn Lighting in a Forest", "autumn_forest"), ("Gentle Snowfall at Dusk", "snowfall_dusk"), ("Hazy Light of a Winter Morning", "winter_hazy"), ("Rain-soaked Reflections in City Lights", "rain_city"), ("Golden Sunlight Streaming through Trees", "trees_golden_sun"), ("Fireflies Lighting up a Summer Night", "fireflies_summer"), ("Glowing Embers from a Forge", "forge_embers"), ("Dim Candlelight in a Gothic Castle", "gothic_castle"), ("Midnight Sky with Bright Starlight", "starlight_midnight"), ("Warm Sunset in a Rural Village", "rural_sunset"), ("Flickering Light in a Haunted House", "haunted_flicker"), ("Desert Sunset with Mirage-like Glow", "desert_mirage"), ("Golden Beams Piercing through Storm Clouds", "storm_beams"), ("Custom", "custom"), ], value="none", elem_classes="radio-group" ) illumination_env_custom = gr.Textbox( label="Custom Illumination Environment", placeholder="e.g., Inside a crystal palace, Underwater volcano, etc.", visible=False ) with gr.Tab("Custom Prompt"): with gr.Accordion("✍️ Custom Prompt (in any language)", open=False): prompt = gr.Textbox( placeholder="Example: Add warm sunset lighting from the right", lines=3 ) with gr.Row(): reset_btn = gr.Button("🔄 Reset") run_btn = gr.Button("✨ Generate", variant="primary") with gr.Accordion("⚙️ Advanced Settings", open=False): seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0) randomize_seed = gr.Checkbox(label="Randomize Seed", value=True) true_guidance_scale = gr.Slider(label="True Guidance Scale", minimum=1.0, maximum=10.0, step=0.1, value=1.0) num_inference_steps = gr.Slider(label="Inference Steps", minimum=1, maximum=40, step=1, value=4) height = gr.Slider(label="Height", minimum=256, maximum=2048, step=8, value=1024) width = gr.Slider(label="Width", minimum=256, maximum=2048, step=8, value=1024) with gr.Column(scale=1): result = gr.Image(label="Output Image", interactive=False) prompt_preview = gr.Textbox(label="Processed Prompt (in Chinese)", interactive=False) create_video_button = gr.Button("🎥 Create Video Between Images", variant="secondary", visible=False) with gr.Group(visible=False) as video_group: video_output = gr.Video(label="Generated Video", show_download_button=True, autoplay=True) inputs = [ image, light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt, seed, randomize_seed, true_guidance_scale, num_inference_steps, height, width, prev_output ] outputs = [result, seed, prompt_preview] # Toggle custom textboxes visibility light_type.change(fn=toggle_custom_textbox, inputs=[light_type], outputs=[light_type_custom], queue=False) light_direction.change(fn=toggle_custom_textbox, inputs=[light_direction], outputs=[light_direction_custom], queue=False) light_intensity.change(fn=toggle_custom_textbox, inputs=[light_intensity], outputs=[light_intensity_custom], queue=False) illumination_env.change(fn=toggle_custom_textbox, inputs=[illumination_env], outputs=[illumination_env_custom], queue=False) # Reset behavior reset_btn.click( fn=reset_all, inputs=None, outputs=[light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt, is_reset], queue=False ).then(fn=end_reset, inputs=None, outputs=[is_reset], queue=False) # Manual generation with video button visibility control def infer_and_show_video_button(*args): result_img, result_seed, result_prompt = infer_relight(*args) # Show video button if we have both input and output images show_button = args[0] is not None and result_img is not None return result_img, result_seed, result_prompt, gr.update(visible=show_button) run_event = run_btn.click( fn=infer_and_show_video_button, inputs=inputs, outputs=outputs + [create_video_button] ) # Video creation create_video_button.click( fn=lambda: gr.update(visible=True), outputs=[video_group], api_name=False ).then( fn=create_video_between_images, inputs=[image, result, prompt_preview], outputs=[video_output], api_name=False ) # Examples gr.Examples( examples=[ ["harold.png", "dramatic", "", "side", "", "soft", "", "none", "", "", 0, True, 1.0, 4, 672, 1024], ["distracted.png", "golden_hour", "", "side", "", "strong", "", "none", "", "", 0, True, 1.0, 4, 640, 1024], ["disaster.jpg", "moonlight", "", "front", "", "medium", "", "neon_city", "", "", 0, True, 1.0, 4, 640, 1024], ], inputs=[image, light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt, seed, randomize_seed, true_guidance_scale, num_inference_steps, height, width], outputs=outputs, fn=infer_relight, cache_examples="lazy", elem_id="examples" ) # Image upload triggers dimension update and control reset image.upload( fn=update_dimensions_on_upload, inputs=[image], outputs=[width, height] ).then( fn=reset_all, inputs=None, outputs=[light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt, is_reset], queue=False ).then( fn=end_reset, inputs=None, outputs=[is_reset], queue=False ) # Live updates - only trigger on non-custom radio selections def maybe_infer(is_reset, progress=gr.Progress(track_tqdm=True), *args): if is_reset: return gr.update(), gr.update(), gr.update(), gr.update() else: result_img, result_seed, result_prompt = infer_relight(*args) # Show video button if we have both input and output show_button = args[0] is not None and result_img is not None return result_img, result_seed, result_prompt, gr.update(visible=show_button) control_inputs = [ image, light_type, light_type_custom, light_direction, light_direction_custom, light_intensity, light_intensity_custom, illumination_env, illumination_env_custom, prompt, seed, randomize_seed, true_guidance_scale, num_inference_steps, height, width, prev_output ] control_inputs_with_flag = [is_reset] + control_inputs # Only trigger live updates when selecting non-custom options def should_trigger_infer(choice): return choice != "custom" for control in [light_type, light_direction, light_intensity, illumination_env]: control.input( fn=lambda choice, is_reset_val, *args, progress=gr.Progress(track_tqdm=True): maybe_infer(is_reset_val, progress, *args) if should_trigger_infer(choice) else (gr.update(), gr.update(), gr.update(), gr.update()), inputs=[control, is_reset] + control_inputs, # Pass control separately, then is_reset, then the rest outputs=outputs + [create_video_button] ) run_event.then(lambda img, *_: img, inputs=[result], outputs=[prev_output]) demo.launch()