ComfyUI-SPEED Dials Up Image Generation To Nearly Double The Speed

ComfyUI-SPEED is a new custom node for ComfyUI that speeds up image generation by using a technique called Spectral Progressive Diffusion. The node can nearly double sampling speed by starting at a lower resolution and growing the image detail as the denoising process continues. It was tested specifically with the Anima model and integrates with the SamplerCustomAdvanced node.

The implementation comes from a developer known as ruwwww, who described the project as “vibecoded.” They built the node based on the SPEED research paper by Howard Xiao and colleagues to offer a practical, training-free acceleration method. The work gives local ComfyUI users a straightforward way to cut generation time with a drop-in sampler.

Faster sampling with simple setup

People who run image generation on consumer GPUs will find this tool useful for faster iterations. Privacy‑conscious professionals and serious hobbyists can keep their workflows local while cutting the time spent waiting for results. Small agencies that rely on local AI for work can also benefit from the speedup without switching to cloud services.

Developer notes and limitations

The developer warns that the node can produce visible artifacts, especially from upsampling, and output quality varies by model and prompt. Testing showed that using torch.compile did not help performance and actually made sampling slower in this implementation. For a balanced speed boost on Anima, the creator recommends setting transition_1 to 0.8 and transition_2 to 0.7, which yields around a 1.4× speedup.

"It’s a pretty vibecoded implementation, so don’t expect polished engineering or faithful implementation given official code isn't out yet, but it does the thing." — Source: Reddit