Add files using upload-large-folder tool

.gitattributes

@@ -585,3 +585,15 @@ python_env/lib/site-packages/pandas/tests/indexing/__pycache__/test_loc.cpython-
 python_env/lib/site-packages/pandas/tests/io/__pycache__/test_sql.cpython-310.pyc filter=lfs diff=lfs merge=lfs -text
 python_env/lib/site-packages/pandas/tests/tools/__pycache__/test_to_datetime.cpython-310.pyc filter=lfs diff=lfs merge=lfs -text
 WebView2Loader.dll filter=lfs diff=lfs merge=lfs -text
+assets/architecture.webp filter=lfs diff=lfs merge=lfs -text
+assets/DMDR.webp filter=lfs diff=lfs merge=lfs -text
+assets/image_arena_all.jpg filter=lfs diff=lfs merge=lfs -text
+assets/decoupled-dmd.webp filter=lfs diff=lfs merge=lfs -text
+assets/leaderboard.png filter=lfs diff=lfs merge=lfs -text
+assets/image_arena_os.jpg filter=lfs diff=lfs merge=lfs -text
+assets/showcase.jpg filter=lfs diff=lfs merge=lfs -text
+assets/training_pipeline.jpg filter=lfs diff=lfs merge=lfs -text
+assets/reasoning.png filter=lfs diff=lfs merge=lfs -text
+assets/showcase_editing.png filter=lfs diff=lfs merge=lfs -text
+assets/showcase_rendering.png filter=lfs diff=lfs merge=lfs -text
+assets/Z-Image-Gallery.pdf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -12,7 +12,9 @@
 <a href="https://arxiv.org/abs/2511.22699" target="_blank"><img src="https://img.shields.io/badge/Report-b5212f.svg?logo=arxiv" height="21px"></a>
 
 
-Welcome to the official repository for the Z-Image（造相）project!
+这是一个基于通义Z-Image-Turbo 的项目，主要是为了满足低显存的设备能够快速出图，同时也增加了一些常用功能，图生图，局部重绘之类的。
+
+This is a project based on Tongyi Z-Image-Turbo. It is mainly designed to enable devices with low video memory to generate images quickly. Meanwhile, some common functions such as image-to-image generation and local redrawing have also been added.
 
 </div>
 

Z-Image-Launcher.exe

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:591d859487add54f3e5431e097989b713870904fa824ffec88fd3cefd87278fa
-size 1339705
+oid sha256:d5673674af1ea04b8424396da60676f91d9e3b4690d61f071fafd9b0705a76be
+size 1413433

app.py

@@ -0,0 +1,1040 @@
+import os
+import time  # 【新增】引入time模块用于退出延迟
+
+# 必须处于文件最顶端：环境配置
+os.environ["DIFFUSERS_USE_PEFT_BACKEND"] = "1"
+os.environ["HF_HUB_DISABLE_SYMLINKS_WARNING"] = "1"
+os.environ["HF_HUB_OFFLINE"] = "1"
+
+import sys
+import torch
+import psutil
+import random
+import re
+import uuid
+import gc
+from datetime import datetime
+from PIL import Image, ImageFilter, ImageOps, ImageEnhance
+
+# 配置基础路径
+current_dir = os.path.dirname(os.path.abspath(__file__))
+if current_dir not in sys.path:
+    sys.path.append(current_dir)
+
+# 目录配置
+DEFAULT_MODEL_PATH = os.path.join(current_dir, "ckpts", "Z-Image-Turbo")
+LORA_ROOT = os.path.join(current_dir, "lora")
+OUTPUT_ROOT = os.path.join(current_dir, "outputs")
+MOD_VAE_DIR = os.path.join(current_dir, "Mod", "vae")
+MOD_TRANS_DIR = os.path.join(current_dir, "Mod", "transformer")
+for p in [LORA_ROOT, OUTPUT_ROOT, MOD_VAE_DIR, MOD_TRANS_DIR]:
+    os.makedirs(p, exist_ok=True)
+
+try:
+    import gradio as gr
+    from diffusers import ZImagePipeline, ZImageImg2ImgPipeline, AutoencoderKL
+    from safetensors.torch import load_file
+except ImportError as e:
+    print(f"❌ 核心库导入失败: {e}")
+    sys.exit(1)
+
+# ==========================================
+# 设备探测与硬件报告
+# ==========================================
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+DTYPE = torch.bfloat16 if DEVICE == "cuda" else torch.float32
+is_interrupted = False
+
+print("\n" + "="*50)
+if DEVICE == "cuda":
+    GPU_NAME = torch.cuda.get_device_name(0)
+    TOTAL_VRAM = torch.cuda.get_device_properties(0).total_memory
+    print(f"✅ 运行模式: [ GPU ]")
+    print(f"核心型号: {GPU_NAME}")
+    print(f"显存总量: {TOTAL_VRAM/1024**3:.2f} GB")
+else:
+    TOTAL_VRAM = 0
+    print(f"⚠️ 运行模式: [ CPU ]")
+print("="*50 + "\n")
+
+# ==========================================
+# 显存与工具函数
+# ==========================================
+def get_vram_info():
+    if DEVICE == "cuda":
+        reserved = torch.cuda.memory_reserved(0)
+        allocated = torch.cuda.memory_allocated(0)
+        usage_pct = (reserved / TOTAL_VRAM) * 100 if TOTAL_VRAM > 0 else 0
+        vram_str = (
+            f"显存占用: {usage_pct:.1f}% "
+            f"({reserved/1024**3:.2f}GB / {TOTAL_VRAM/1024**3:.2f}GB)"
+        )
+    else:
+        usage_pct = 0
+        vram_str = "显存占用: CPU 模式"
+
+    mem = psutil.virtual_memory()
+    ram_str = (
+        f"内存占用: {mem.percent:.1f}% "
+        f"({(mem.total - mem.available)/1024**3:.2f}GB / {mem.total/1024**3:.2f}GB)"
+    )
+    status = f"{vram_str} ｜ {ram_str}"
+    return usage_pct, status
+
+
+def auto_flush_vram(threshold=90):
+    usage_pct, _ = get_vram_info()
+    if usage_pct > threshold:
+        gc.collect()
+        torch.cuda.empty_cache()
+        return True
+    return False
+
+def scan_lora_files():
+    if not os.path.exists(LORA_ROOT): return []
+    return sorted([f for f in os.listdir(LORA_ROOT) if f.lower().endswith(".safetensors")])
+
+def scan_model_items(base_path):
+    if not os.path.exists(base_path): return []
+    items = []
+    for f in os.listdir(base_path):
+        full_path = os.path.join(base_path, f)
+        if os.path.isdir(full_path):
+            items.append(f)
+        elif f.lower().endswith((".safetensors", ".bin", ".pt")):
+            items.append(f)
+    return sorted(items)
+
+# ==========================================
+# 全局 LoRA 文件列表 (启动时扫描)
+# ==========================================
+LORA_FILES = scan_lora_files()
+print(f"🔍 已检测到 {len(LORA_FILES)} 个 LoRA 文件，正在生成独立控件...")
+if len(LORA_FILES) > 30:
+    print("⚠️ 警告: LoRA 数量较多，生成界面可能需要几秒钟...")
+
+# ==========================================
+# 模型管理器 (修改版：支持性能模式切换)
+# ==========================================
+class ModelManager:
+    def __init__(self):
+        self.pipe = None 
+        self.current_state = {
+            "mode": None,      
+            "t_choice": None,  
+            "v_choice": None,
+            "perf_mode": None  # 【新增】记录当前性能模式
+        }
+        self.current_loras = []
+        self.current_weights_map = {} 
+
+    def _clear_pipeline(self):
+        if self.pipe is not None:
+            print(f"🧹 正在销毁旧管道以释放显存...")
+            try:
+                self.pipe.unload_lora_weights()
+            except:
+                pass
+            del self.pipe
+            self.pipe = None
+        if hasattr(sys, 'last_traceback'):
+            del sys.last_traceback
+        for _ in range(3):
+            gc.collect()
+        torch.cuda.empty_cache()
+        torch.cuda.ipc_collect()
+        if DEVICE == "cuda":
+            res = torch.cuda.memory_reserved(0) / 1024**3
+            print(f"✨ 显存已深度清理，当前占用: {res:.2f} GB")
+
+    def _init_pipeline_base(self, mode):
+        if mode == 'txt':
+            print("🚀 初始化基础 Pipeline (文生图)...")
+            return ZImagePipeline.from_pretrained(DEFAULT_MODEL_PATH, torch_dtype=DTYPE, local_files_only=True)
+        else:
+            print("🚀 初始化基础 Pipeline (图生图)...")
+            return ZImageImg2ImgPipeline.from_pretrained(DEFAULT_MODEL_PATH, torch_dtype=DTYPE, local_files_only=True)
+
+    def _inject_components(self, pipe, t_choice, v_choice):
+        if t_choice != "default":
+            t_path = os.path.join(MOD_TRANS_DIR, t_choice)
+            if os.path.isfile(t_path):
+                print(f"📦 载入 Transformer: {t_choice}")
+                state_dict = load_file(t_path, device="cpu")
+                processed = {}
+                prefix = "model.diffusion_model."
+                for k, v in state_dict.items():
+                    new_k = k[len(prefix):] if k.startswith(prefix) else k
+                    processed[new_k] = v.to(DTYPE)
+                pipe.transformer.load_state_dict(processed, strict=False, assign=True)
+                del state_dict, processed, v
+                gc.collect()
+
+        if v_choice != "default":
+            vae_path = os.path.join(MOD_VAE_DIR, v_choice)
+            print(f"📦 载入 VAE: {v_choice}")
+            if os.path.isfile(vae_path):
+                pipe.vae = AutoencoderKL.from_single_file(vae_path, torch_dtype=DTYPE)
+            else:
+                pipe.vae = AutoencoderKL.from_pretrained(vae_path, torch_dtype=DTYPE)
+        return pipe
+
+    def _apply_loras(self, pipe, selected_loras, weights_map):
+        if self.current_loras == selected_loras and self.current_weights_map == weights_map:
+            return
+
+        print("🎸 正在配置 LoRA (独立权重模式)...")
+        try:
+            pipe.unload_lora_weights()
+        except Exception:
+            pass
+
+        if not selected_loras:
+            self.current_loras = []
+            self.current_weights_map = {}
+            return
+
+        active_adapters = []
+        adapter_weights = []
+
+        for lora_file in selected_loras:
+            adapter_name = re.sub(r"[^a-zA-Z0-9_]", "_", os.path.splitext(lora_file)[0])
+            weight = weights_map.get(lora_file, 1.0)
+            
+            try:
+                pipe.load_lora_weights(LORA_ROOT, weight_name=lora_file, adapter_name=adapter_name)
+                active_adapters.append(adapter_name)
+                adapter_weights.append(weight)
+            except Exception as e:
+                print(f"⚠️ LoRA {lora_file} 加载失败: {e}")
+
+        if active_adapters:
+            pipe.set_adapters(active_adapters, adapter_weights=adapter_weights)
+        
+        self.current_loras = list(selected_loras)
+        self.current_weights_map = dict(weights_map)
+
+    def get_pipeline(self, t_choice, v_choice, selected_loras, weights_map, mode='txt', perf_mode="高端机 (显存>=20GB)"):
+        # 判断是否为低端机模式
+        is_low_vram = (perf_mode == "低端机 (显存优化)")
+
+        # 检查是否需要重建模型 (包含性能模式变更的检查)
+        need_rebuild = (
+            self.pipe is None or
+            self.current_state["mode"] != mode or
+            self.current_state["t_choice"] != t_choice or
+            self.current_state["v_choice"] != v_choice or
+            self.current_state["perf_mode"] != perf_mode  # 【新增】性能模式变更需重建
+        )
+
+        if need_rebuild:
+            self._clear_pipeline() 
+            try:
+                temp_pipe = self._init_pipeline_base(mode)
+                temp_pipe = self._inject_components(temp_pipe, t_choice, v_choice)
+                
+                # 【核心修改】根据选择应用不同的显存优化策略
+                if DEVICE == "cuda":
+                    if is_low_vram:
+                        temp_pipe.enable_sequential_cpu_offload()
+                        print("  [System] 已启用低显存优化模式")
+                    else:
+                        temp_pipe.to("cuda")
+                        print("  [System] 已启用高端机模式")
+
+                self.pipe = temp_pipe
+                
+                self.current_state = {
+                    "mode": mode,
+                    "t_choice": t_choice,
+                    "v_choice": v_choice,
+                    "perf_mode": perf_mode
+                }
+                self.current_loras = [] 
+                self.current_weights_map = {}
+                
+            except Exception as e:
+                self._clear_pipeline()
+                raise gr.Error(f"模型加载崩溃: {str(e)}\n请检查显存或模型文件。")
+
+        self._apply_loras(self.pipe, selected_loras, weights_map)
+        return self.pipe
+
+manager = ModelManager()
+
+# ==========================================
+# 进度回调
+# ==========================================
+def make_progress_callback(progress, total_steps, refresh_interval=2):
+    def _callback(pipe, step, timestep, callback_kwargs):
+        global is_interrupted
+        if is_interrupted:
+            raise gr.Error("🛑 任务已手动停止")
+        step_idx = step + 1
+        frac = step_idx / total_steps
+        status_suffix = ""
+        if step_idx % refresh_interval == 0 or step_idx == total_steps:
+            _, mem_status = get_vram_info()
+            status_suffix = f"\n{mem_status}"
+        progress(frac, desc=f"Diffusion Step {step_idx}/{total_steps}{status_suffix}")
+        return callback_kwargs
+    return _callback
+
+# ==========================================
+# 核心逻辑 (解析独立控件传入的参数)
+# ==========================================
+def process_lora_inputs(lora_checks, lora_weights):
+    selected = []
+    weights_map = {}
+    for i, fname in enumerate(LORA_FILES):
+        if i < len(lora_checks) and lora_checks[i]:
+            selected.append(fname)
+            if i < len(lora_weights):
+                weights_map[fname] = lora_weights[i]
+            else:
+                weights_map[fname] = 1.0
+    return selected, weights_map
+
+# 【新增】LoRA 刷新函数
+def refresh_lora_list():
+    global LORA_FILES
+    LORA_FILES = scan_lora_files()
+    count = len(LORA_FILES)
+    print(f"🔄 LoRA 刷新完成，当前检测到 {count} 个文件。")
+    # 提示用户刷新网页以看到新控件（因为 Gradio 静态布局限制）
+    msg = f"✅ 扫描完成！检测到 **{count}** 个 LoRA 文件。\n\n*(注：若新增了文件，请**刷新浏览器页面 (F5)** 以加载新的勾选框和滑块)*"
+    return gr.update(value=msg)
+
+# 【新增】更新 Prompt UI 的辅助函数
+def update_prompt_ui_base(prompt, *lora_ui_args):
+    num_loras = len(LORA_FILES)
+    if num_loras == 0:
+        return prompt
+
+    checks = lora_ui_args[:num_loras]
+    weights = lora_ui_args[num_loras:num_loras*2]
+
+    clean_p = re.sub(r"\s*<lora:[^>]+>", "", prompt or "").strip()
+    
+    new_tags = []
+    for i, fname in enumerate(LORA_FILES):
+        if i < len(checks) and checks[i]:
+            w = weights[i] if i < len(weights) else 1.0
+            name = os.path.splitext(fname)[0]
+            alpha_str = f"{w:.2f}".rstrip("0").rstrip(".")
+            new_tags.append(f"<lora:{name}:{alpha_str}>")
+    
+    if new_tags:
+        return f"{clean_p} {' '.join(new_tags)}"
+    else:
+        return clean_p
+
+# 【修复】使用 *args 接收参数，避免 Gradio 传参顺序问题
+def run_inference(*args):
+    global is_interrupted
+    is_interrupted = False
+    
+    # 解析参数顺序
+    # [prompt, checks(N), weights(N), t, v, perf_mode, w, h, steps, cfg, seed, random, batch, vram_th]
+    idx = 0
+    prompt = args[idx]; idx += 1
+    num_loras = len(LORA_FILES)
+    lora_checks = args[idx : idx+num_loras]; idx += num_loras
+    lora_weights = args[idx : idx+num_loras]; idx += num_loras
+    
+    t_choice = args[idx]; idx += 1
+    v_choice = args[idx]; idx += 1
+    perf_mode = args[idx]; idx += 1 # 【新增】性能模式
+    
+    w = args[idx]; idx += 1
+    h = args[idx]; idx += 1
+    steps = args[idx]; idx += 1
+    cfg = args[idx]; idx += 1
+    seed = args[idx]; idx += 1
+    is_random = args[idx]; idx += 1
+    batch_size = args[idx]; idx += 1
+    vram_threshold = args[idx]; idx += 1
+
+    auto_flush_vram(vram_threshold)
+    clean_w = (int(w) // 16) * 16
+    clean_h = (int(h) // 16) * 16
+    
+    selected_loras, weights_map = process_lora_inputs(lora_checks, lora_weights)
+    
+    # 构建最终 Prompt
+    if selected_loras:
+        tags = []
+        for f in selected_loras:
+            w_val = weights_map.get(f, 1.0)
+            name = os.path.splitext(f)[0]
+            tags.append(f"<lora:{name}:{w_val:.2f}>")
+        clean_p = re.sub(r"\s*<lora:[^>]+>", "", prompt or "").strip()
+        final_prompt = f"{clean_p} {' '.join(tags)}"
+    else:
+        final_prompt = prompt
+
+    try:
+        # 【修改】传入 perf_mode
+        pipe = manager.get_pipeline(t_choice, v_choice, selected_loras, weights_map, mode='txt', perf_mode=perf_mode)
+    except Exception as e:
+        raise gr.Error(f"模型加载失败: {str(e)}")
+
+    if is_random: seed = random.randint(0, 2**32 - 1)
+    generator = torch.Generator(DEVICE).manual_seed(int(seed))
+
+    date_folder = datetime.now().strftime("%Y-%m-%d")
+    save_dir = os.path.join(OUTPUT_ROOT, date_folder)
+    os.makedirs(save_dir, exist_ok=True)
+
+    results_images = []
+    progress = gr.Progress()
+
+    try:
+        print(f"🔥 任务启动 | 图片分辨率: {clean_w}x{clean_h} | 种子: {seed}")
+        step_callback = make_progress_callback(progress, int(steps))
+
+        for i in range(int(batch_size)):
+            if is_interrupted: break
+            output = pipe(
+                prompt=final_prompt,
+                width=clean_w,
+                height=clean_h,
+                num_inference_steps=int(steps),
+                guidance_scale=float(cfg),
+                generator=generator,
+                callback_on_step_end=step_callback
+            ).images[0]
+
+            filename = f"{datetime.now().strftime('%H%M%S')}_{uuid.uuid4().hex[:4]}.png"
+            path = os.path.join(save_dir, filename)
+            output.save(path)
+            results_images.append(output)
+            _, current_status = get_vram_info()
+            yield results_images, seed, current_status
+
+    except Exception as e:
+        if "任务已手动停止" in str(e):
+            print("🛑 任务已停止")
+        else:
+            import traceback
+            traceback.print_exc()
+            raise gr.Error(f"生成中断: {str(e)}")
+    finally:
+        # del pipe
+        auto_flush_vram(vram_threshold)
+
+# 【修复】图生图
+def run_img2img(*args, progress=gr.Progress()):
+    global is_interrupted
+    is_interrupted = False
+    
+    # [input_image, prompt, checks(N), weights(N), t, v, perf_mode, ...fixed...]
+    idx = 0
+    input_image = args[idx]; idx += 1
+    prompt = args[idx]; idx += 1
+    
+    num_loras = len(LORA_FILES)
+    lora_checks = args[idx : idx+num_loras]; idx += num_loras
+    lora_weights = args[idx : idx+num_loras]; idx += num_loras
+    
+    t_choice = args[idx]; idx += 1
+    v_choice = args[idx]; idx += 1
+    perf_mode = args[idx]; idx += 1 # 【新增】性能模式
+    
+    output_width = args[idx]; idx += 1
+    output_height = args[idx]; idx += 1
+    strength = args[idx]; idx += 1
+    steps = args[idx]; idx += 1
+    cfg = args[idx]; idx += 1
+    seed = args[idx]; idx += 1
+    is_random = args[idx]; idx += 1
+    batch_size = args[idx]; idx += 1
+    vram_threshold = args[idx]; idx += 1
+
+    if input_image is None:
+        raise gr.Error("❌ 请先上传图片")
+        
+    auto_flush_vram(vram_threshold)
+    selected_loras, weights_map = process_lora_inputs(lora_checks, lora_weights)
+
+    if selected_loras:
+        tags = []
+        for f in selected_loras:
+            w_val = weights_map.get(f, 1.0)
+            name = os.path.splitext(f)[0]
+            tags.append(f"<lora:{name}:{w_val:.2f}>")
+        clean_p = re.sub(r"\s*<lora:[^>]+>", "", prompt or "").strip()
+        final_prompt = f"{clean_p} {' '.join(tags)}"
+    else:
+        final_prompt = prompt
+
+    if output_width == 0 or output_height == 0:
+        orig_w, orig_h = input_image.size
+        aspect = orig_w / orig_h
+        target_size = 1024
+        if aspect > 1:
+            target_w, target_h = target_size, max(512, int(target_size / aspect))
+        else:
+            target_h, target_w = target_size, max(512, int(target_size * aspect))
+        target_w = (target_w // 16) * 16
+        target_h = (target_h // 16) * 16
+    else:
+        target_w = (int(output_width) // 16) * 16
+        target_h = (int(output_height) // 16) * 16
+
+    input_image = input_image.convert("RGB").resize((target_w, target_h))
+
+    if is_random: seed = random.randint(0, 2**32 - 1)
+    generator = torch.Generator(DEVICE).manual_seed(int(seed))
+
+    date_folder = datetime.now().strftime("%Y-%m-%d")
+    save_dir = os.path.join(OUTPUT_ROOT, date_folder)
+    os.makedirs(save_dir, exist_ok=True)
+
+    results = []
+    pipe = None
+    
+    try:
+        # 【修改】传入 perf_mode
+        pipe = manager.get_pipeline(t_choice, v_choice, selected_loras, weights_map, mode='img', perf_mode=perf_mode)
+
+        for i in progress.tqdm(range(int(batch_size)), desc="图生图生成中"):
+            if is_interrupted: break
+            torch.cuda.ipc_collect()
+            step_callback = make_progress_callback(progress, int(steps))
+
+            output = pipe(
+                prompt=final_prompt,
+                image=input_image,
+                strength=float(strength),
+                num_inference_steps=int(steps),
+                guidance_scale=0.0,
+                generator=generator,
+                callback_on_step_end=step_callback
+            ).images[0]
+
+            filename = f"img2img_{datetime.now().strftime('%H%M%S')}_{uuid.uuid4().hex[:4]}.png"
+            path = os.path.join(save_dir, filename)
+            output.save(path)
+            results.append(path)
+
+    except Exception as e:
+        if "任务已手动停止" in str(e):
+            print("🛑 任务已停止")
+        else:
+            import traceback
+            traceback.print_exc()
+            raise gr.Error(f"生成中断: {str(e)}")
+    finally:
+        del pipe
+        auto_flush_vram(vram_threshold)
+        _, current_status = get_vram_info()
+
+    return results, seed, current_status
+
+# 【修复】融合图
+def run_fusion_img(*args, progress=gr.Progress()):
+    global is_interrupted
+    is_interrupted = False
+    
+    # [image1, image2, prompt, checks(N), weights(N), t, v, perf_mode, ...fixed...]
+    idx = 0
+    image1 = args[idx]; idx += 1
+    image2 = args[idx]; idx += 1
+    prompt = args[idx]; idx += 1
+    
+    num_loras = len(LORA_FILES)
+    lora_checks = args[idx : idx+num_loras]; idx += num_loras
+    lora_weights = args[idx : idx+num_loras]; idx += num_loras
+    
+    t_choice = args[idx]; idx += 1
+    v_choice = args[idx]; idx += 1
+    perf_mode = args[idx]; idx += 1 # 【新增】性能模式
+    
+    output_width = args[idx]; idx += 1
+    output_height = args[idx]; idx += 1
+    blend_strength = args[idx]; idx += 1
+    strength = args[idx]; idx += 1
+    steps = args[idx]; idx += 1
+    cfg = args[idx]; idx += 1
+    seed = args[idx]; idx += 1
+    is_random = args[idx]; idx += 1
+    batch_size = args[idx]; idx += 1
+    vram_threshold = args[idx]; idx += 1
+
+    if image1 is None or image2 is None:
+        raise gr.Error("❌ 请上传两张参考图片")
+        
+    auto_flush_vram(vram_threshold)
+    selected_loras, weights_map = process_lora_inputs(lora_checks, lora_weights)
+
+    if selected_loras:
+        tags = []
+        for f in selected_loras:
+            w_val = weights_map.get(f, 1.0)
+            name = os.path.splitext(f)[0]
+            tags.append(f"<lora:{name}:{w_val:.2f}>")
+        clean_p = re.sub(r"\s*<lora:[^>]+>", "", prompt or "").strip()
+        final_prompt = f"{clean_p} {' '.join(tags)}"
+    else:
+        final_prompt = prompt
+
+    if output_width == 0 or output_height == 0:
+        orig_w, orig_h = image1.size
+        aspect = orig_w / orig_h
+        target_size = 1024
+        if aspect > 1:
+            target_w, target_h = target_size, max(512, int(target_size / aspect))
+        else:
+            target_h, target_w = target_size, max(512, int(target_size * aspect))
+        target_w = (target_w // 16) * 16
+        target_h = (target_h // 16) * 16
+    else:
+        target_w = (int(output_width) // 16) * 16
+        target_h = (int(output_height) // 16) * 16
+
+    image1 = image1.convert("RGB").resize((target_w, target_h))
+    image2 = image2.convert("RGB").resize((target_w, target_h))
+    blended_image = Image.blend(image1, image2, float(blend_strength))
+
+    if is_random: seed = random.randint(0, 2**32 - 1)
+    generator = torch.Generator(DEVICE).manual_seed(int(seed))
+
+    date_folder = datetime.now().strftime("%Y-%m-%d")
+    save_dir = os.path.join(OUTPUT_ROOT, date_folder)
+    os.makedirs(save_dir, exist_ok=True)
+
+    results = []
+    pipe = None
+    
+    try:
+        # 【修改】传入 perf_mode
+        pipe = manager.get_pipeline(t_choice, v_choice, selected_loras, weights_map, mode='img', perf_mode=perf_mode)
+
+        for i in progress.tqdm(range(int(batch_size)), desc="融合生成中"):
+            if is_interrupted: break
+            torch.cuda.ipc_collect()
+            step_callback = make_progress_callback(progress, int(steps))
+
+            output = pipe(
+                prompt=final_prompt,
+                image=blended_image,
+                strength=float(strength),
+                num_inference_steps=int(steps),
+                guidance_scale=0.0,
+                generator=generator,
+                callback_on_step_end=step_callback
+            ).images[0]
+
+            filename = f"fusion_{datetime.now().strftime('%H%M%S')}_{uuid.uuid4().hex[:4]}.png"
+            path = os.path.join(save_dir, filename)
+            output.save(path)
+            results.append(path)
+
+    except Exception as e:
+        if "任务已手动停止" in str(e):
+            print("🛑 任务已停止")
+        else:
+            import traceback
+            traceback.print_exc()
+            raise gr.Error(f"生成中断: {str(e)}")
+    finally:
+        del pipe
+        auto_flush_vram(vram_threshold)
+        _, current_status = get_vram_info()
+
+    return results, seed, current_status
+
+# ==========================================
+# UI 界面
+# ==========================================
+js_kill_window = """
+function() {
+    setTimeout(function(){ window.close(); }, 1000);
+    document.body.innerHTML = '<div style="display:flex;justify-content:center;align-items:center;height:100vh;background:#000;color:#fff;font-family:sans-serif;"><h1>🚫 系统已关闭，请直接关闭此标签页</h1></div>';
+    document.body.style.backgroundColor = "black";
+    document.body.style.overflow = "hidden";
+    return [];
+}
+"""
+
+def kill_system_process():
+    print("🛑 正在执行一键退出程序...")
+    try:
+        os.system("taskkill /F /IM Z-Image-Launcher.exe")
+    except Exception:
+        pass
+    time.sleep(1)
+    try:
+        os.system("taskkill /F /IM python.exe")
+    except Exception:
+        pass
+    sys.exit(0)
+
+# 【新增】根据显存��动判断默认模式
+DEFAULT_PERF_MODE = "低端机 (显存优化)" if TOTAL_VRAM < 20 * 1024**3 else "高端机 (显存>=20GB)"
+
+with gr.Blocks(title="造相 Z-Image Pro Studio | 作者: ") as demo:
+
+    print('\n' + '!'*60)
+    print('  本软件由 Leewheel 免费分享，严禁售卖！')
+    print('!'*60 + '\n')
+    gr.Warning('本软件由 Leewheel 免费分享。如果你是付费购买，你被骗了！', duration=20)
+    with gr.Row(elem_id="header_row"):
+        gr.Markdown("# 🎨 造相 Z-Image Pro Studio | 作者:  Leewheel(V1.00C)")
+        exit_btn = gr.Button("❌ 一键退出系统", variant="stop", scale=0, min_width=150)
+        
+    vram_info_display = gr.Markdown("显存状态加载中...")
+
+    with gr.Tabs():
+        # --- 文成图 ---
+        with gr.Tab("文成图"):
+            with gr.Row():
+                with gr.Column(scale=4):
+                    prompt_input = gr.Textbox(label="Prompt", lines=4)
+                    manual_flush_btn = gr.Button("🧹 清理显存", size="sm", variant="secondary")
+                    vram_threshold_slider = gr.Slider(50, 98, 90, step=1, label="自动清理阈值 (%)")
+                    
+                    # 【修改】LoRA 区域添加刷新按钮
+                    with gr.Accordion("LoRA 权重设置 (每个 LoRA 独立调节)", open=False):
+                        txt_lora_checks = []
+                        txt_lora_sliders = []
+                        
+                        # 【新增】LoRA 刷新按钮和状态提示
+                        with gr.Row():
+                            txt_refresh_lora_btn = gr.Button("🔄 刷新 LoRA 文件列表", size="sm")
+                            txt_lora_info_md = gr.Markdown("")
+                        
+                        if not LORA_FILES:
+                            gr.Markdown("*未检测到 LoRA 文件*")
+                        else:
+                            for fname in LORA_FILES:
+                                with gr.Row():
+                                    chk = gr.Checkbox(label=fname, value=False, scale=1, container=False)
+                                    sld = gr.Slider(0, 2.0, 1.0, step=0.05, label="权重", scale=4)
+                                    txt_lora_checks.append(chk)
+                                    txt_lora_sliders.append(sld)
+
+                    with gr.Accordion("模型设置", open=True):
+                        refresh_models_btn = gr.Button("🔄 刷新底模/VAE", size="sm")
+                        t_drop = gr.Dropdown(label="Transformer", choices=["default"] + scan_model_items(MOD_TRANS_DIR), value="default")
+                        v_drop = gr.Dropdown(label="VAE", choices=["default"] + scan_model_items(MOD_VAE_DIR), value="default")
+                        
+                        # 【新增】性能模式选项
+                        perf_mode_radio = gr.Radio(
+                            choices=["高端机 (显存>=20GB)", "低端机 (显存优化)"],
+                            value=DEFAULT_PERF_MODE,
+                            label="性能模式"
+                        )
+                        
+                        with gr.Row():
+                            width_s = gr.Slider(512, 2048, 1024, step=16, label="宽 (16倍数)")
+                            height_s = gr.Slider(512, 2048, 1024, step=16, label="高 (16倍数)")
+                        step_s = gr.Slider(1, 50, 8, label="步数")
+                        cfg_s = gr.Slider(0, 10, 0, label="CFG")
+                        batch_s = gr.Slider(1, 32, 1, step=1, label="生成张数")
+                        seed_n = gr.Number(label="种子", value=42, precision=0)
+                        random_c = gr.Checkbox(label="随机种子", value=True)
+
+                    with gr.Row():
+                        run_btn = gr.Button("🚀 开始生成", variant="primary", size="lg")
+                        stop_btn = gr.Button("🛑 停止生成", variant="stop", size="lg", interactive=False)
+
+                with gr.Column(scale=6):
+                    res_gallery = gr.Gallery(label="输出结果", columns=2, height="80vh")
+                    res_seed = gr.Number(label="种子", interactive=False)
+                    vram_info_display = gr.Markdown("显存状态加载中...")
+
+        # --- 图片编辑 ---
+        with gr.Tab("图片编辑"):
+            with gr.Row():
+                with gr.Column():
+                    image_input = gr.Image(label="上传图片", type="pil")
+                    with gr.Group():
+                        rotate_angle = gr.Slider(-360, 360, 0, step=1, label="旋转角度 (度)")
+                        crop_x = gr.Slider(0, 100, 0, step=1, label="裁剪 X (%)")
+                        crop_y = gr.Slider(0, 100, 0, step=1, label="裁剪 Y (%)")
+                        crop_width = gr.Slider(0, 100, 100, step=1, label="裁剪宽度 (%)")
+                        crop_height = gr.Slider(0, 100, 100, step=1, label="裁剪高度 (%)")
+                        flip_horizontal = gr.Checkbox(label="水平翻转")
+                        flip_vertical = gr.Checkbox(label="垂直翻转")
+                    edit_btn = gr.Button("开始编辑", variant="primary")
+                with gr.Column():
+                    edited_image_output = gr.Image(label="编辑后的图片", type="pil")
+                    with gr.Group():
+                        apply_filter = gr.Dropdown(["模糊", "轮廓", "细节", "边缘增强", "更多边缘增强", "浮雕", "查找边缘", "锐化", "平滑", "更多平滑"], label="应用滤镜")
+                        brightness = gr.Slider(-100, 100, 0, step=1, label="亮度调整 (%)")
+                        contrast = gr.Slider(-100, 100, 0, step=1, label="对比度调整 (%)")
+                        saturation = gr.Slider(-100, 100, 0, step=1, label="饱和度调整 (%)")
+
+            def edit_image(image, angle, x, y, width, height, hflip, vflip, filter, brightness, contrast, saturation):
+                if image is None: return None
+                if angle != 0: image = image.rotate(angle, expand=True)
+                if x or y or width < 100 or height < 100:
+                    original_width, original_height = image.size
+                    left = int(original_width * x / 100)
+                    top = int(original_height * y / 100)
+                    right = int(original_width * (x + width) / 100)
+                    bottom = int(original_height * (y + height) / 100)
+                    image = image.crop((left, top, right, bottom))
+                if hflip: image = ImageOps.mirror(image)
+                if vflip: image = ImageOps.flip(image)
+                if filter:
+                    filter_map = {
+                        "模糊": ImageFilter.BLUR, "轮廓": ImageFilter.CONTOUR, "细节": ImageFilter.DETAIL,
+                        "边缘增强": ImageFilter.EDGE_ENHANCE, "更多边缘增强": ImageFilter.EDGE_ENHANCE_MORE,
+                        "浮雕": ImageFilter.EMBOSS, "查找边缘": ImageFilter.FIND_EDGES,
+                        "锐化": ImageFilter.SHARPEN, "平滑": ImageFilter.SMOOTH, "更多平滑": ImageFilter.SMOOTH_MORE
+                    }
+                    filter_func = filter_map.get(filter)
+                    if filter_func: image = image.filter(filter_func)
+                if brightness != 0:
+                    enhancer = ImageEnhance.Brightness(image)
+                    image = enhancer.enhance(1 + brightness / 100)
+                if contrast != 0:
+                    enhancer = ImageEnhance.Contrast(image)
+                    image = enhancer.enhance(1 + contrast / 100)
+                if saturation != 0:
+                    enhancer = ImageEnhance.Color(image)
+                    image = enhancer.enhance(1 + saturation / 100)
+                return image
+
+            edit_btn.click(
+                fn=edit_image,
+                inputs=[image_input, rotate_angle, crop_x, crop_y, crop_width, crop_height, flip_horizontal, flip_vertical, apply_filter, brightness, contrast, saturation],
+                outputs=edited_image_output
+            )
+
+        # --- 图生图 ---
+        with gr.Tab("图生图"):
+            with gr.Row():
+                with gr.Column(scale=4):
+                    with gr.Group():
+                        img2img_input = gr.Image(label="上传参考图", type="pil")
+                        img2img_prompt = gr.Textbox(label="Prompt (推荐)", lines=2, placeholder="描述你想要生成的画面...")
+                        img2img_flush = gr.Button("🧹 清理显存", size="sm", variant="secondary")
+                        
+                        # 【修改】图生图独立控件，添加刷新
+                        with gr.Accordion("LoRA 权重设置 (独立调节)", open=False):
+                            i2i_lora_checks = []
+                            i2i_lora_sliders = []
+                            
+                            with gr.Row():
+                                i2i_refresh_lora_btn = gr.Button("🔄 刷新 LoRA 文件列表", size="sm")
+                                i2i_lora_info_md = gr.Markdown("")
+
+                            if not LORA_FILES:
+                                gr.Markdown("*未检测到 LoRA 文件*")
+                            else:
+                                for fname in LORA_FILES:
+                                    with gr.Row():
+                                        chk = gr.Checkbox(label=fname, value=False, scale=1, container=False)
+                                        sld = gr.Slider(0, 2.0, 1.0, step=0.05, label="权重", scale=4)
+                                        i2i_lora_checks.append(chk)
+                                        i2i_lora_sliders.append(sld)
+
+                    with gr.Accordion("模型与参数", open=True):
+                        img2img_refresh_models = gr.Button("🔄 ��新底模/VAE", size="sm")
+                        img2img_t_drop = gr.Dropdown(label="Transformer", choices=["default"] + scan_model_items(MOD_TRANS_DIR), value="default")
+                        img2img_v_drop = gr.Dropdown(label="VAE", choices=["default"] + scan_model_items(MOD_VAE_DIR), value="default")
+                        
+                        # 【新增】图生图性能模式
+                        img2img_perf_mode = gr.Radio(
+                            choices=["高端机 (显存>=20GB)", "低端机 (显存优化)"],
+                            value=DEFAULT_PERF_MODE,
+                            label="性能模式"
+                        )
+                        
+                        with gr.Row():
+                            img2img_width_s = gr.Slider(0, 2048, 0, step=16, label="输出宽 (0=自动保持比例)")
+                            img2img_height_s = gr.Slider(0, 2048, 0, step=16, label="输出高 (0=自动保持比例)")
+                        gr.Markdown("**提示：** 宽高都为0时自动保持上传图比例并接近1024；手动设置大于512时生效")
+                        img2img_strength = gr.Slider(0.0, 1.0, 0.75, step=0.01, label="重绘强度")
+                        img2img_steps = gr.Slider(1, 100, 12, step=1, label="步数")
+                        img2img_cfg = gr.Number(value=0.0, label="CFG（Turbo模型固定为0.0）", interactive=False)
+                        img2img_batch = gr.Slider(1, 8, 1, step=1, label="张数")
+                        img2img_seed = gr.Number(label="种子", value=42, precision=0)
+                        img2img_random = gr.Checkbox(label="随机种子", value=True)
+                    with gr.Row():
+                        img2img_run_btn = gr.Button("🚀 生成", variant="primary", size="lg")
+                        img2img_stop_btn = gr.Button("🛑 停止", variant="stop", size="lg", interactive=False)
+                with gr.Column(scale=6):
+                    img2img_gallery = gr.Gallery(label="图生图结果", columns=2, height="80vh")
+                    img2img_res_seed = gr.Number(label="种子", interactive=False)
+
+        # --- 融合图 ---
+        with gr.Tab("融合图"):
+            gr.Markdown("**融合2张图片**：图片1提供主要结构/姿势，图片2提供细节/脸部/风格。")
+            with gr.Row():
+                with gr.Column(scale=4):
+                    with gr.Group():
+                        fusion_input1 = gr.Image(label="图片1（主结构/姿势）", type="pil")
+                        fusion_input2 = gr.Image(label="图片2（细节/脸部/风格）", type="pil")
+                        fusion_prompt = gr.Textbox(label="融合描述 Prompt", lines=3)
+                        fusion_flush = gr.Button("🧹 清理显存", size="sm", variant="secondary")
+                        
+                        # 【修改】融合图独立控件，添加刷新
+                        with gr.Accordion("LoRA 权重设置 (独立调节)", open=False):
+                            fusion_lora_checks = []
+                            fusion_lora_sliders = []
+                            
+                            with gr.Row():
+                                fusion_refresh_lora_btn = gr.Button("🔄 刷新 LoRA 文件列表", size="sm")
+                                fusion_lora_info_md = gr.Markdown("")
+
+                            if not LORA_FILES:
+                                gr.Markdown("*未检测到 LoRA 文件*")
+                            else:
+                                for fname in LORA_FILES:
+                                    with gr.Row():
+                                        chk = gr.Checkbox(label=fname, value=False, scale=1, container=False)
+                                        sld = gr.Slider(0, 2.0, 1.0, step=0.05, label="权重", scale=4)
+                                        fusion_lora_checks.append(chk)
+                                        fusion_lora_sliders.append(sld)
+
+                    with gr.Accordion("模型与参数", open=True):
+                        fusion_refresh_models = gr.Button("🔄 刷新底模/VAE", size="sm")
+                        fusion_t_drop = gr.Dropdown(label="Transformer", choices=["default"] + scan_model_items(MOD_TRANS_DIR), value="default")
+                        fusion_v_drop = gr.Dropdown(label="VAE", choices=["default"] + scan_model_items(MOD_VAE_DIR), value="default")
+                        
+                        # 【新增】融合图性能模式
+                        fusion_perf_mode = gr.Radio(
+                            choices=["高端机 (显存>=20GB)", "低端机 (显存优化)"],
+                            value=DEFAULT_PERF_MODE,
+                            label="性能模式"
+                        )
+                        
+                        with gr.Row():
+                            fusion_width_s = gr.Slider(0, 2048, 0, step=16, label="输出宽 (0=自动保持比例)")
+                            fusion_height_s = gr.Slider(0, 2048, 0, step=16, label="输出高 (0=自动保持比例)")
+                        gr.Markdown("**提示：** 宽高都为0时自动保持图片1比例并接近1024")
+                        with gr.Row():
+                            fusion_blend = gr.Slider(0.0, 1.0, 0.5, step=0.05, label="图片2混合强度 (0=全用图片1, 1=全用图片2)")
+                            fusion_strength = gr.Slider(0.0, 1.0, 0.7, step=0.05, label="重绘强度 (越高变化越大)")
+                        fusion_steps = gr.Slider(1, 100, 15, step=1, label="步数")
+                        fusion_cfg = gr.Number(value=0.0, label="CFG（固定为0.0）", interactive=False)
+                        fusion_batch = gr.Slider(1, 8, 1, step=1, label="张数")
+                        fusion_seed = gr.Number(label="种子", value=42, precision=0)
+                        fusion_random = gr.Checkbox(label="随机种子", value=True)
+                    with gr.Row():
+                        fusion_run_btn = gr.Button("🚀 开始融合", variant="primary", size="lg")
+                        fusion_stop_btn = gr.Button("🛑 停止", variant="stop", size="lg", interactive=False)
+                with gr.Column(scale=6):
+                    fusion_gallery = gr.Gallery(label="融合结果", columns=2, height="80vh")
+                    fusion_res_seed = gr.Number(label="种子", interactive=False)
+
+    # -----------------------
+    # UI状态函数
+    # -----------------------
+    def ui_to_running():
+        return gr.update(interactive=False), gr.update(interactive=True)
+
+    def ui_to_idle():
+        return gr.update(interactive=True), gr.update(interactive=False)
+
+    def trigger_interrupt():
+        global is_interrupted
+        is_interrupted = True
+        return "🛑 正在强制中断..."
+
+    # -----------------------
+    # 按钮事件绑定
+    # -----------------------
+    
+    # 退出按钮
+    exit_btn.click(fn=kill_system_process, js=js_kill_window)
+
+    # 文生图
+    refresh_models_btn.click(
+        fn=lambda: (
+            gr.update(choices=["default"] + scan_model_items(MOD_TRANS_DIR)),
+            gr.update(choices=["default"] + scan_model_items(MOD_VAE_DIR))
+        ),
+        outputs=[t_drop, v_drop]
+    )
+    
+    # 【新增】文生图 LoRA 刷新
+    txt_refresh_lora_btn.click(fn=refresh_lora_list, outputs=txt_lora_info_md)
+    
+    manual_flush_btn.click(
+        fn=lambda: (gc.collect(), torch.cuda.empty_cache(), get_vram_info()[1])[2],
+        outputs=vram_info_display
+    )
+
+    txt_ui_inputs = [prompt_input] + txt_lora_checks + txt_lora_sliders
+    for c in txt_lora_checks + txt_lora_sliders:
+        c.change(fn=update_prompt_ui_base, inputs=txt_ui_inputs, outputs=prompt_input)
+
+    inference_event = run_btn.click(
+        fn=ui_to_running, 
+        outputs=[run_btn, stop_btn]
+    ).then(
+        fn=run_inference,
+        # 【修改】输入增加 perf_mode_radio
+        inputs=txt_ui_inputs + [t_drop, v_drop, perf_mode_radio, width_s, height_s, step_s, cfg_s, seed_n, random_c, batch_s, vram_threshold_slider],
+        outputs=[res_gallery, res_seed, vram_info_display]
+    ).then(
+        fn=ui_to_idle,
+        outputs=[run_btn, stop_btn]
+    )
+
+    stop_btn.click(
+        fn=trigger_interrupt,
+        outputs=vram_info_display
+    ).then(
+        fn=ui_to_idle,
+        outputs=[run_btn, stop_btn],
+        cancels=[inference_event]
+    )
+    
+    # 图生图
+    def refresh_all_models_img():
+        return gr.update(choices=["default"] + scan_model_items(MOD_TRANS_DIR)), gr.update(choices=["default"] + scan_model_items(MOD_VAE_DIR))
+    img2img_refresh_models.click(fn=refresh_all_models_img, outputs=[img2img_t_drop, img2img_v_drop])
+    
+    # 【新增】图生图 LoRA 刷新
+    i2i_refresh_lora_btn.click(fn=refresh_lora_list, outputs=i2i_lora_info_md)
+    
+    img2img_flush.click(fn=lambda: (gc.collect(), torch.cuda.empty_cache(), get_vram_info()[1])[2], outputs=vram_info_display)
+
+    i2i_ui_inputs = [img2img_prompt] + i2i_lora_checks + i2i_lora_sliders
+    for c in i2i_lora_checks + i2i_lora_sliders:
+        c.change(fn=update_prompt_ui_base, inputs=i2i_ui_inputs, outputs=img2img_prompt)
+
+    img2img_event = img2img_run_btn.click(fn=ui_to_running, outputs=[img2img_run_btn, img2img_stop_btn])\
+        .then(fn=run_img2img,
+              # 【修改】输入增加 img2img_perf_mode
+              inputs=[img2img_input, img2img_prompt] + i2i_lora_checks + i2i_lora_sliders + 
+                      [img2img_t_drop, img2img_v_drop, img2img_perf_mode, img2img_width_s, img2img_height_s,
+                       img2img_strength, img2img_steps, img2img_cfg, img2img_seed, img2img_random, img2img_batch, vram_threshold_slider],
+              outputs=[img2img_gallery, img2img_res_seed, vram_info_display])\
+        .then(fn=ui_to_idle, outputs=[img2img_run_btn, img2img_stop_btn])
+
+    img2img_stop_btn.click(fn=trigger_interrupt, outputs=vram_info_display).then(fn=ui_to_idle, outputs=[img2img_run_btn, img2img_stop_btn], cancels=[img2img_event])
+
+    # 融合图
+    fusion_refresh_models.click(fn=refresh_all_models_img, outputs=[fusion_t_drop, fusion_v_drop])
+    
+    # 【新增】融合图 LoRA 刷新
+    fusion_refresh_lora_btn.click(fn=refresh_lora_list, outputs=fusion_lora_info_md)
+    
+    fusion_flush.click(fn=lambda: (gc.collect(), torch.cuda.empty_cache(), get_vram_info()[1])[2], outputs=vram_info_display)
+
+    fusion_ui_inputs = [fusion_prompt] + fusion_lora_checks + fusion_lora_sliders
+    for c in fusion_lora_checks + fusion_lora_sliders:
+        c.change(fn=update_prompt_ui_base, inputs=fusion_ui_inputs, outputs=fusion_prompt)
+
+    fusion_event = fusion_run_btn.click(fn=ui_to_running, outputs=[fusion_run_btn, fusion_stop_btn])\
+        .then(fn=run_fusion_img,
+              # 【修改】输入增加 fusion_perf_mode
+              inputs=[fusion_input1, fusion_input2, fusion_prompt] + fusion_lora_checks + fusion_lora_sliders + 
+                      [fusion_t_drop, fusion_v_drop, fusion_perf_mode, fusion_width_s, fusion_height_s,
+                       fusion_blend, fusion_strength, fusion_steps, fusion_cfg, 
+                       fusion_seed, fusion_random, fusion_batch, vram_threshold_slider],
+              outputs=[fusion_gallery, fusion_res_seed, vram_info_display])\
+        .then(fn=ui_to_idle, outputs=[fusion_run_btn, fusion_stop_btn])
+
+    fusion_stop_btn.click(fn=trigger_interrupt, outputs=vram_info_display).then(fn=ui_to_idle, outputs=[fusion_run_btn, fusion_stop_btn], cancels=[fusion_event])
+
+if __name__ == "__main__":
+    demo.launch(share=False, inbrowser=True)

assets/Z-Image-Gallery.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f9895b3246d2547bac74bbe0be975da500eaae93f2cad4248ad3281786b1ac6
+size 15767436