Please note that the free download is a trial version and includes all features except multi-region repaint.

The workflow has high requirements for the graphics card, and it can run stably when tested on the 4090, and the graphics card below this specification is not sure whether it will run normally.

——

The full version introduces the features：

1、ChangeX is a multi-area redrawing workflow developed based on the SDXL model. It enables simultaneous redrawing of single or multiple areas under specified modes, supporting the simultaneous replacement of backgrounds, faces, clothing, hairstyles, objects, and more, along with foreground generation and additional interactive features... It further incorporates the FLUX model for post-processing optimization, significantly enhancing the image quality.

2、ChangeX offers a comprehensive set of function groups, process switches, and parameter control nodes, allowing for rapid customization and formulation of dedicated workflows tailored to specific needs.

3、The operational interface is categorized through a card-based distribution, arranged sequentially from left to right following the process flow, ensuring clearer and more logical operations. Additionally, it provides detailed function groups for finer control over individual aspects, facilitating a more precise handling of details.

___

Update Content  (1218):

The main purpose of the update is to enrich the functions and simplify the process:

1. Added the upload node of the region mask;

2. You can use a separate Lora for style control for the region;

3. Added the function that you can select the position of "粘贴还原（mean's paste and restore）" in the running process. For example, you can use Flux to optimize and then paste the reserved area of the image back, or paste the reserved area back and then use Flux to optimize it.

4. Modified the image batch processing logic to avoid insufficient VRAM;

5. Removed the node that controls the expression because it will cause the image to be blurry;

6. Some interface optimizations, node and parameter adjustments.

Update Content  (1209):

1. The V150 version simplifies the operation of the reference image part, and now only uses the way of drawing masks to crop the content of the picture, which can greatly reduce the number of processing nodes to improve efficiency;

2. Optimize the function and layout of the operation interface to make it more reasonable;

3. Use Florence2 to generate more accurate inverse words;

4. The logic of secondary sampling has been optimized, and the test shows that the graph effect is more stable.

5. Shorten the processing path of some nodes, improve efficiency, and fix some bugs.

Update Content (1101):

1.Change to using the Flux GGUF model, and also switch the upscaling module to the Flux + Ultimate SD Upscale method. This will significantly reduce the VRAM requirements while still ensuring high-quality rendering;

2. Some process optimization.

Update Content (1016):

1、Adjusted the logic for generating images. The first image on the right side of the comparison is now changed to a preview of the sampling result (the image generated based on the number of sampling times used), and the second image is changed to the final output image effect. Additionally, the process sequence for restoring the post has been moved to the end, meaning that the restoration of the post will be performed last in the process.

2、Removed the mask locking function due to numerous unresolved issues associated with this function node. Therefore, it has been deleted.

3、The node switch for using the MaskEditor to draw masks for reference images has been changed to use masks for cropping. The reference images can now be cropped using a mask framing method.

4、Fixed some issues and optimized the process.

Update Contents (1014):

1.By default, the Inpaint model is disabled during the first sampling, which in most cases can enhance image quality and generate more reasonable backgrounds.

2.Added a batch image processing function that allows for continuous area redrawing on multiple images at once.

3.Fixed some bugs.

Update Contents (1012):

1、Added an efficiency node to enhance logical processing. Place the MultiNumericJudges_node.py node file in the ComfyUI/custom_nodes/ directory.

2、The area segmentation feature now includes a mask locking function. Place the lock_mask.py node file in the ComfyUI/custom_nodes/ directory.

3、The background depth map control group has been enhanced with a line art strengthening function.

4、Improvements have been made to some interfaces, process efficiency, and image quality.

——————

工作流介绍：

1、ChangeX是基于SDXL模型开发的多区域重绘工作流，可以对单个或多个区域同时进行指定模式下的重绘，支持同时更换背景、人脸、衣服、发型、物品等等，以及生成前景功能和更多的玩法……并结合了FLUX模型进行后期优化，进一步提升图片质感。

2、ChangeX提供了完善的功能群组、流程开关以及参数控制节点，可以根据需求快速组合从而制定专用流程。

3、操作界面使用卡片分布的方式进行了分类，依照流程顺序从左到右依次排列，操作更加清晰合理；并提供了细分功能群组，方便对细节进一步把控。

___

更新内容（1218）：

更新主要目的是丰富功能同时简化流程：

1. 增加了区域蒙版上传节点；
2. 可以为区域使用单独Lora进行风格控制；
3. 增加了可以选择“回贴还原”在跑图流程中的位置的功能，例如：可以先使用Flux优化再回贴还原，或者先回贴还原再使用Flux进行优化等等。
4. 修改了批量处理逻辑，可以避免爆显存的情况；
5. 把之前的InstantID+Flux的换脸方式改成在Flux优化时利用PuLID进行换脸，这样可以简化流程并缩短处理时间同时还可以获得不错的效果；删除了效果不佳的表情控制节点；

6、一些界面优化，节点以及参数调整等等。

更新内容（1209）：

1、V150版本简化对参考图部分的操作，现在仅使用绘制蒙版的方式来裁剪画面内容，这样可以大幅减少处理节点的数量以提升效率；

2、优化操作界面的功能及布局，使其更加合理；

3、使用Florence2生成更加准确的反推词；

4、优化了二次采样的逻辑，测试显示出图效果更加稳定。

5、缩短一些节点的处理路径，提升效率，以及修复了一些错误。

更新内容（1101）：

1、改成使用Flux GGUF模型，同时放大模块也换成Flux + Ultimate SD Upscale的方式，这样对VRAM的要求会降低不少同时也能保证出图质量；

2、一些流程优化。

更新内容（1016）：

1.调整生成图片逻辑，对比图右边第一张改为采样结果预览（根据所使用的采样次数所生成的图片），第二张改为最终输出的图片效果；同时，把回贴还原流程顺序调到最后，意思是在流程最后才进行回贴还原。

2.去掉蒙版锁定功能，该功能节点存在过多问题暂时无法解决，所以删除。

3.参考图原使用MaskEditor绘制蒙版的节点开关改为使用蒙版进行裁剪，可以使用蒙版画框的方式去对参考图进行裁剪。

4.一些问题修复以及流程优化。

更新内容（1014）：

1.默认首次采样关闭Inpaint模型，大部份情况下能提升图片质量，生成的背景更合理些；

2.增加批量处理图片功能，可一次性连续对多张图片进行区域重绘；

3.修复一些bug。

更新内容（1012）：

1.增加了效率节点以提升逻辑处理，把 MultiNumericJudges_node.py 节点⽂件放到 ComfyUI/custom_nodes/ ⽬录下

2.区域分割增加了蒙版锁定功能，该功能还在测试中，目前只能应用在单个区域，需要把节点文件 lock_mask.py 节点文件放到ComfyUI/custom_nodes/ ⽬录下

3.背景深度图控制群组增加了线稿加强功能；

4.一些界面、流程效率优化，以及画质改善。

——————

How-to video tutorials

视频教程：

https://youtu.be/xp3DqkB6vEA?si=4-eIEl8MtMdQgTed

https://www.bilibili.com/video/BV1T3xPeVEAb/

——————

The full version is accessible through sponsorship. You can sponsor as much as you like, depending on your personal preferences. Your support is my greatest motivation. Thank you!

For the time being, sponsorship can only be made via PayPal:

https://paypal.me/xiser88

If you have already sponsored, please reply to the email to notify me. I'll send you the workflow file as soon as I see the reply.

My email address: [email protected]

——————

国内用户请通过QQ联系我：3861103314