what is comfyui-universal-seamless-tiles fr?

This project adds two custom nodes to ComfyUI, a visual tool for building AI image generation workflows, that make images tileable, meaning they can be repeated side by side without a visible seam where the edges meet. The older trick for making tileable images only worked on certain older AI model designs. Newer image generation models such as Flux and Wan do not respond to that old trick at all, since they are built differently under the hood. This project's approach works across both the older model types and these newer ones, so it does not depend on which specific AI model you are using. It achieves this in two steps. First, one node subtly shifts the image data during the generation process itself, in a way that gradually decreases as generation nears completion, which prevents the model from settling on a fixed, visible edge. Second, a separate node adjusts how the final image is decoded from that internal data so the pixels at the true edges wrap around cleanly, removing any remaining seam. Neither node changes the original AI model or decoder permanently, they work on a copy, so using them will not affect your other image generations. To use the tool, you connect your loaded model through the first node before your normal sampling step, and connect your VAE, the component responsible for turning generated data into a final image, through the second node before the final decode step. Both nodes need to be set to the same tiling setting, and you can check whether the tiling worked by shifting the finished image by half its width and height and looking for a visible seam. The README notes that the technique works best with more generation steps, so very fast few-step models may need extra steps for the cleanest results, and that this method does not currently work together with tools like ControlNet. It has been tested with several popular current image generation models and is released under the MIT license.