wtf is comfyui-speed?

ComfyUI-SPEED is an add-on for AI image and video generation tools that makes the picture creation process faster without noticeably lowering quality. It is built as an unofficial version of a technique called SPEED, described in a research paper, and it plugs into three different popular AI art generation programs: ComfyUI, Forge Neo, and SwarmUI. AI image generators normally build a picture gradually, refining random noise into a finished image over many steps, all at the image's full resolution the whole time. This tool changes that process so early steps work on a smaller, lower resolution version of the image and only expand up to full resolution as the picture becomes clearer. Because the early steps do less work per step, the whole generation finishes faster. The README shows an example where a test image generated about a third faster using this method compared to the standard approach, with a similar looking result. Installing it means placing the project's folder into the right extensions or custom nodes directory for whichever of the three supported programs you use, then restarting that program. Once installed, you enable it through a settings panel or by connecting it into the generation workflow, and it works alongside whichever underlying sampling method you already had selected. The tool comes with several adjustable settings, such as which resolution steps to pass through, how sensitive the timing of those steps is, and which mathematical method is used to shrink and expand the image data. Certain features do not work well together with this tool, such as inpainting, where you paint over and regenerate part of an existing image, and some advanced sampler types are not supported. The underlying technique and research come from a separate official project by other researchers, released under the BSD 3-Clause license. This repository is a community reimplementation, and it points users back to the original project for full details on licensing and authorship.