shootthesound/comfyui-kms

ComfyUI-KMS is a custom node for ComfyUI, a popular interface for running image-generation AI models. The node is called KSampler Multi-Choice, and it solves a common frustration: choosing the right "seed." A seed is the random starting point that determines what an image will look like. Normally, if you want to compare several options, you have to generate each image fully, which costs time and computing power. This node changes that workflow. It runs a short, partial render for several seeds in a row, just enough steps to see the rough composition of each image. Then it pauses and shows those partial renders as small clickable thumbnails directly on the node itself. You look at the thumbnails, pick the one you like best, and only that chosen seed gets rendered to completion. The others are discarded unless you decide you want them later. A key feature is that the work from that initial partial run is not wasted. The finished image continues from exactly where the short probe stopped. With deterministic samplers, the result is identical to what you would get if you had just rendered that seed from scratch the normal way. This makes browsing cheap. The developer notes that looking at eight seeds this way costs around 22 total model steps, compared to 64 steps for rendering eight full images. The node works with different model types. For fast, distilled models, a one or two step probe is often enough to see the composition. For traditional multi-step models, you might set the probe to run for about a third of the total steps so the preview is clear enough to judge. After you pick a seed, the other thumbnails stay on screen. If you change your mind, you can click another candidate and it will render too, picking up from its own cached stopping point. An example workflow is included to help you get started. The full README is longer than what was shown.