FLUX 2 KLEIN 9B（智能替换） - 一键精准替换复杂场景的物体/人物 一键整合包下载

无言***

FLUX 2 KLEIN 9B（智能替换） - 一键精准替换物体/人物 一键整合包下载

FLUX 2 KLEIN 9B（Smart Character Replacement）是一个基于Klein 9B模型的智能物体和人物替换工具，可以用来精确地替换那些复杂或拥挤的图像中的人物或物体，同时不破坏图像的整体质量。特别适合想在复杂图片里精准换人、又不想把整张图搞崩的用户。

相比之前的复杂人物或物体替换，传统模式下需要手动绘制蒙版，将需要替换的人物或物体手动画出来，今天分享的软件利用最新的 SAM 3 遮罩技术自动检测主体，用占位符替换所选的人物或物体，保留图像的其余部分，同时可防止在多次编辑过程中出现的像素偏移问题。



主要特点

精准替换：能准确识别并只替换指定人物（比如“把穿黄衣服的女人换成参考图里的人”），其他背景、其他人、灯光、姿势基本不动。
使用遮罩（Mask）技术：虽然 Flux 模型本身可以不用遮罩直接改图，但作者强烈推荐用遮罩。因为多次修改后不用遮罩容易出现“画崩”（脸变歪、画质下降、像素漂移）。用了遮罩后可以反复编辑同一张图，质量更稳定，适合认真做图的人。
纯原生节点：整个工作流只用 ComfyUI 自带的节点，不依赖乱七八糟的第三方插件，打开就用，比较稳定。
自动辅助：包含 SAM 3 自动遮罩功能，能帮你更快地选中要替换的人物。



应用领域

人物写真/换脸：把自己的脸换到各种场景的照片里。
群像修改：多人照片里只换其中一个人的脸或身体。
影视/商业图像后期：需要精确控制某个角色，但不想破坏整个画面的情况。
创意设计：快速把参考角色塞进复杂场景中做概念图、插画等。
迭代编辑：同一张图可以多次修改不同部分，适合精修用户。




使用教程：（建议N卡，显存8G起，支持50系显卡）

整合包包含所需所有节点，下载主程序和模型（ComfyUI文件夹），解压主程序一键包，将ComfyUI文件夹移动到主程序目录下即可，详情参考下方目录结构

WebUI：启动后，自动进入WebUI，上传主图和需要替换的图像，输入提示词和分割目标提示词，生成即可。

ComfyUI工作流：
双击启动，浏览器输入 http://127.0.0.1:8188/ 进入页面后，点击左侧的 工作流程，选择对应的工作流，上传主图和需要替换的图像，输入提示词和分割目标提示词，运行即可。


关于分割目标提示词：
使用最新的Sam3.1分割模型，可以简单描述图中需要分割的物体或人物，从衣服颜色、头发颜色以及位置等区分

“the woman in yellow dress”
“穿黄色裙子的女人”

“the first man”  “第一个人”

“the second man”  “第二个人”

你还可以更详细地描述复杂的场景，从而提高描述的准确性


主提示词也尽量详细描述，比如

“Replace the woman in the yellow dress with the character in Image 2.”
“把穿黄色裙子的女人换成图片 2 中的那个角色。”

软件目录结构

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 unet/
│   │      └── flux-2-klein-9b-Q4_K_M.gguf
│   ├── 📂 text_encoders/
│   │      └── Qwen3-8B-Q4_K_M.gguf
│   ├── 📂 vae/
│   │      └── flux2-vae.safetensors
📂 deepface/
......



下载地址：
迅雷云盘：https://pan.xunlei.com/s/VOtRZhpI8d13blQofaXkl4A5A1?pwd=n2s6

夸克网盘：

🔒付费内容

游客， 上上宾会员 可免费下载该资源，点此开通上上宾 免费下载全站99%的付费资源。或单独支付 20碎银 下载该资源

立即购买

百度网盘：

🔒付费内容

游客， 上上宾会员 可免费下载该资源，点此开通上上宾 免费下载全站99%的付费资源。或单独支付 40碎银 下载该资源

立即购买

hyl***

试了好多次，每次都出现gradio.exceptions.Error: '生成超时'错误，请问是什么原因？

无言***

hylsry 发表于 2026-5-27 11:54
试了好多次，每次都出现gradio.exceptions.Error: '生成超时'错误，请问是什么原因？
...

完整的 报错信息贴出来

hyl***

无言以对 发表于 2026-5-27 12:03
完整的 报错信息贴出来

[12:13:12] [Launcher] 启动 ComfyUI...
[12:13:12] [Launcher] 等待 ComfyUI 启动...
[12:13:14] [ComfyUI] WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.
[12:13:14] [ComfyUI] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4']}
[12:13:14] [ComfyUI] Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
[12:13:14] [ComfyUI] Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []}
[12:13:14] [ComfyUI] Checkpoint files will always be loaded safely.
[12:13:14] [ComfyUI] Total VRAM 16379 MB, total RAM 32602 MB
[12:13:14] [ComfyUI] pytorch version: 2.7.1+cu128
[12:13:14] [ComfyUI] Set vram state to: NORMAL_VRAM
[12:13:14] [ComfyUI] Device: cuda:0 NVIDIA GeForce RTX 4060 Ti : cudaMallocAsync
[12:13:14] [ComfyUI] Using async weight offloading with 2 streams
[12:13:14] [ComfyUI] Enabled pinned memory 13040.0
[12:13:15] [ComfyUI] Using pytorch attention
[12:13:17] [ComfyUI] Unsupported Pytorch detected. DynamicVRAM support requires Pytorch version 2.8 or later. Falling back to legacy ModelPatcher. VRAM estimates may be unreliable especially on Windows
[12:13:17] [ComfyUI] Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr  8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
[12:13:17] [ComfyUI] ComfyUI version: 0.21.1
[12:13:17] [ComfyUI] comfy-aimdo version: 0.3.0
[12:13:17] [ComfyUI] comfy-kitchen version: 0.2.8
[12:13:17] [ComfyUI] ComfyUI frontend version: 1.43.18
[12:13:17] [ComfyUI] [Prompt Server] web root: F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\comfyui_frontend_package\static
[12:13:17] [ComfyUI] Asset seeder disabled
[12:13:17] [ComfyUI] ComfyUI-GGUF: Partial torch compile only, consider updating pytorch
[12:13:17] [ComfyUI] Import times for custom nodes:
[12:13:17] [ComfyUI] 0.0 seconds: F:\ComfyUI_Klein-Replacement\ComfyUI\custom_nodes\websocket_image_save.py
[12:13:17] [ComfyUI] 0.0 seconds: F:\ComfyUI_Klein-Replacement\ComfyUI\custom_nodes\ComfyUI-GGUF
[12:13:17] [ComfyUI] 0.1 seconds: F:\ComfyUI_Klein-Replacement\ComfyUI\custom_nodes\ComfyUI-to-Python-Extension
[12:13:17] [ComfyUI] Context impl SQLiteImpl.
[12:13:17] [ComfyUI] Will assume non-transactional DDL.
[12:13:18] [ComfyUI] Starting server
[12:13:18] [ComfyUI] To see the GUI go to: http://127.0.0.1:8188
[12:13:18] [Launcher] ComfyUI 已就绪!
[12:13:18] [Launcher] 启动 Gradio WebUI...
[12:13:20] [Gradio] ComfyUI 连接成功
[12:13:20] [Gradio] * Running on local URL:  http://0.0.0.0:7860
======================================================================
[12:13:23] [Launcher] 所有服务已启动!
[12:13:23] [Launcher] ComfyUI API: http://127.0.0.1:8188
[12:13:23] [Launcher] Gradio WebUI: http://localhost:7860
[12:13:23] [Launcher] 按 Ctrl+C 停止所有服务
======================================================================
[12:13:24] [Gradio] * To create a public link, set `share=True` in `launch()`.
[12:18:51] [ComfyUI] got prompt
[12:18:51] [ComfyUI] Using pytorch attention in VAE
[12:18:51] [ComfyUI] Using pytorch attention in VAE
[12:18:51] [ComfyUI] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[12:18:51] [ComfyUI] model weight dtype torch.float16, manual cast: None
[12:18:51] [ComfyUI] model_type FLOW
[12:18:52] [ComfyUI] WARNING: No VAE weights detected, VAE not initalized.
[12:18:52] [ComfyUI] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[12:18:52] [ComfyUI] Requested to load SAM3ClipModelWrapper
[12:18:52] [ComfyUI] loaded completely; 13703.80 MB usable, 675.18 MB loaded, full load: True
[12:18:53] [ComfyUI] Requested to load SAM3
[12:18:53] [ComfyUI] loaded completely; 12980.50 MB usable, 990.14 MB loaded, full load: True
[12:19:01] [ComfyUI] Requested to load AutoencoderKL
[12:19:01] [ComfyUI] 0 models unloaded.
[12:19:01] [ComfyUI] loaded partially; 0.00 MB usable, 0.00 MB loaded, 160.31 MB offloaded, 13.50 MB buffer reserved, lowvram patches: 0
[12:19:06] [ComfyUI] 0 models unloaded.
[12:19:06] [ComfyUI] loaded partially; 0.00 MB usable, 0.00 MB loaded, 160.31 MB offloaded, 13.50 MB buffer reserved, lowvram patches: 0
[12:19:06] [ComfyUI] Warning: Ran out of memory when regular VAE encoding, retrying with tiled VAE encoding.
[12:20:53] [ComfyUI] gguf qtypes: Q6_K (37), F32 (145), Q4_K (217)
[12:20:53] [ComfyUI] Dequantizing token_embd.weight to prevent runtime OOM.
[12:20:54] [ComfyUI] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[12:20:54] [ComfyUI] Requested to load Flux2TEModel_
[12:20:56] [ComfyUI] loaded completely; 13533.36 MB usable, 6829.34 MB loaded, full load: True
[12:20:59] [ComfyUI] 0 models unloaded.
[12:20:59] [ComfyUI] loaded partially; 0.00 MB usable, 0.00 MB loaded, 160.31 MB offloaded, 13.50 MB buffer reserved, lowvram patches: 0
[12:52:30] [Gradio] Traceback (most recent call last):
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\gradio\queueing.py", line 763, in process_events
[12:52:30] [Gradio] response = await route_utils.call_process_api(
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\gradio\route_utils.py", line 354, in call_process_api
[12:52:30] [Gradio] output = await app.get_blocks().process_api(
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\gradio\blocks.py", line 2125, in process_api
[12:52:30] [Gradio] result = await self.call_function(
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\gradio\blocks.py", line 1607, in call_function
[12:52:30] [Gradio] prediction = await anyio.to_thread.run_sync(  # type: ignore
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\anyio\to_thread.py", line 63, in run_sync
[12:52:30] [Gradio] return await get_async_backend().run_sync_in_worker_thread(
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\anyio\_backends\_asyncio.py", line 2502, in run_sync_in_worker_thread
[12:52:30] [Gradio] return await future
[12:52:30] [Gradio] ^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\anyio\_backends\_asyncio.py", line 986, in run
[12:52:30] [Gradio] result = context.run(func, *args)
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "F:\ComfyUI_Klein-Replacement\deepface\Lib\site-packages\gradio\utils.py", line 1066, in wrapper
[12:52:30] [Gradio] response = f(*args, **kwargs)
[12:52:30] [Gradio] ^^^^^^^^^^^^^^^^^^
[12:52:30] [Gradio] File "<frozen app>", line 279, in generate_image
[12:52:30] [Gradio] gradio.exceptions.Error: '生成超时'
[13:27:41] [ComfyUI] 0 models unloaded.
[13:27:41] [ComfyUI] loaded partially; 0.00 MB usable, 0.00 MB loaded, 160.31 MB offloaded, 13.50 MB buffer reserved, lowvram patches: 0
[13:27:42] [ComfyUI] Warning: Ran out of memory when regular VAE encoding, retrying with tiled VAE encoding.
[13:29:14] [ComfyUI] gguf qtypes: F32 (80), Q5_K (36), Q4_K (76), BF16 (9)
[13:29:14] [ComfyUI] model weight dtype torch.bfloat16, manual cast: None
[13:29:14] [ComfyUI] model_type FLUX
[13:29:14] [ComfyUI] Requested to load Flux2
[13:29:15] [ComfyUI] 0 models unloaded.
[13:29:22] [ComfyUI] loaded partially; 0.00 MB usable, 0.00 MB loaded, 5828.02 MB offloaded, 675.00 MB buffer reserved, lowvram patches: 0
[13:29:22] [ComfyUI] ComfyUI found: F:\ComfyUI_Klein-Replacement\ComfyUI
[13:29:22] [ComfyUI] 'F:\ComfyUI_Klein-Replacement\ComfyUI' added to sys.path

无言***

hylsry 发表于 2026-5-27 13:44
[12:13:12] [Launcher] 启动 ComfyUI...
[12:13:12] [Launcher] 等待 ComfyUI 启动...
[12:13:14] [Comfy ...

默认设置了1200s也就是20分钟的等待时间，如果你超出这个时间，会提示。

你的问题看着像是显存/运存 不足导致的，如果显存≤8G，运存就需要32G起，看看是不是符合？

hyl***

无言以对 发表于 2026-5-27 14:38
默认设置了1200s也就是20分钟的等待时间，如果你超出这个时间，会提示。

你的问题看着像是显存/运存 不 ...

感谢回复，我是4060Ti 16G,内存32G，i5-13400F，我再找找别的原因！

无言***

hylsry 发表于 2026-5-27 15:02
感谢回复，我是4060Ti 16G,内存32G，i5-13400F，我再找找别的原因！

你用comfyui模式生成吧

hyl***

无言以对 发表于 2026-5-27 15:05
你用comfyui模式生成吧

好的！