Generate images and videos with SPEED (Spectral Progressive Diffusion), a training-free method that runs diffusion at progressively increasing resolution for a large speedup over full-resolution sampling. Use this skill whenever the user wants to generate an image with FLUX.1-dev, a pixel-space image with PixelGen, or a video with WAN 2.1 — and especially when they mention SPEED, progressive resolution, faster/efficient/accelerated diffusion sampling, spectral noise expansion, timestep alignment, DCT/DWT/FFT resolution transitions, or running latent_image_gen / pixel_image_gen / latent_video_gen. Reach for it for any "generate an image/video efficiently" request here even if the user never says the word "SPEED".
2026-06-11