MakeTileSEGS in FLUX

A node in ComfyUI-ImpactPack that provides more refined adjustment and overlap between masks compared to the commonly used SDUpscale, resulting in better-looking textures. However, due to the higher number of tiles, the time needed to enhance the image is also longer.

The following workflow offers several options:

Keeping Face From Detailer: The face area is not sampled if you don't bypass the group. Face sampling can be affected by the model. Enabling this group prevents the face from being altered. If you want the face to be sampled, you can bypass this group.

UpscaleUsingModel: Using models to enhance image quality will make the image clearer, but may also reduce image details or introduce unnatural elements. If you notice a reduction in details, you can skip this group.

Additional explanation

The default Resize is set to 2400. With this setting, a square image will be divided into 4 tiles, requiring 4 sampling operations. However, with TeaCache and TurboLora, the speed is still acceptable. If you increase the dimensions, the number of tiles will increase. If you're not concerned about generation time and only want a very large image, even upscaling to 8K will add clear details without obvious seams.

Denoise is set to 0.3. Similar to FLUX's Image2Image settings, higher values result in greater detail changes, while lower values keep the image closer to the original.

The tile size in MakeTileSEGS can be adjusted through bbox_size. Note that it includes scaling factor and borders, so this setting actually results in tiles with dimensions of 1565, while FLUX can go up to 2048. Therefore, when you need a large image, you can appropriately increase the bbox_size to reduce the number of tiles. You can refer to SEGSPreview to see how many tiles the image is divided into.

==

RHTools/WF：

https://www.runninghub.ai/post/1903652745956462593