为解决Lora产生的负面影响！新的lora已经在加紧炼制中，尽早给大家呈现最好的流程，如果你有这套流程的优化建议，请给我反馈，如果真的可以提升，我会署上你的名字并感谢你！这是开源社区该有的样子~我也很荣幸成为一名光荣的开源社区奉献着，希望AGI真的可以为设计帮上什么忙

Update:

9/14 18:11（蒙版边缘）

This workflow is an upgrade in quality and automation based on the background replacement in Majic Product_V3. More importantly, it emphasizes the idea of automation. To achieve full automation, I created a new node called "Snap Area" to obtain the adjusted resolution ratio.

With minimal parameter adjustments, you can accomplish the following:

Adaptive Size: Products of various shapes or sizes can automatically adapt to a fixed resolution. For example, if the product occupies a large proportion of the uploaded image, the resolution of the main subject will automatically adjust to scale down after recalculating to "True."

You can simply choose your output to be in landscape or portrait mode with a button click.

There are seven built-in styles to choose from, and you can implement one by entering a number between 0 and 6.

You can still perform mask editing on the original image to define the occluded areas, and it remains fully adaptive regardless of the resolution or whether it's in landscape or portrait mode.

You can adjust the blending percentage of either the original image or the final output image.

本套工作流在majic product_V3 前 背 景 替换的基础上升级了质量和自动化，但更重要的是自动化思路。为了实现完全自动化，我制作了一个新的节点“Snap Area”来获取化整后的分辨率占比。

你可以在调控极少量参数的情况下完成以下内容

1.自适应大小，让各种形状或大小的产品可以自适应到固定分辨率，例如如果你上传的图像中产品占比较大，那再计算为“真”后会自适应缩小其主体分辨率。

2.你可以通过简单的按钮选择你的产出结果为横/竖屏

3.有7种内置风格供你选择，你只需要手动输入 0 - 6 其中之一，即可实现。

4.依旧可以在原图上进行遮罩编辑来绘制出遮挡区域，且完全不受分辨率或横竖屏影像，依旧自适应。

5.你可以通过百分比来调控原图或最终生成图像的融合度

Snap Area nodes

https://github.com/SS-snap/Snap-Processing.git

LoRa Model Download link

https://civitai.com/models/741380/majicproductv4?modelVersionId=829081

https://www.liblib.art/modelinfo/11dd147ed9be4bd29cbbc517ab26b0f9?from=personal_page

Checkpoints

https://civitai.com/models/133005/juggernaut-xl

controlnet

https://huggingface.co/ckpt/controlnet-sdxl-1.0/blob/main/sai_xl_depth_256lora.safetensors

BrushNet-fp16 model：

https://huggingface.co/streamize/brushnet-sdxl/tree/main

Put it under this path：

ComfyUI-aki-v1.2\models\inpaint\brushnet

ic-light model:

https://huggingface.co/lllyasviel/ic-light/tree/main

put it under this path：

ComfyUI-aki-v1.2\models\unet