What noise painting is trying to do
Noise painting uses hand-crafted adversarial inputs and incongruent prompts to push diffusion into unexpected visual territory.
Lynn Cole / 标签
A thread through the work.
Noise painting uses hand-crafted adversarial inputs and incongruent prompts to push diffusion into unexpected visual territory.
The idea is that you start with a hand painted image or collage, but that you design it in such a way as that it messes up the classification and denoising systems in a diffusion model.
A wedding photograph, a discordant line drawing, CFG 16, and a render that aged the groom and ate the bride.
The layers and source material inside Document 228: a blotter designed to displace and confuse the eye and the machine.
Noise painting began with FaceApp experiments and moved through visual diffusion into sound. Multi-tasking brings those experimental processes together in an album.
An early noise-painting walkthrough, from the source composition through depth corrections, compositing, prompt experiments and the finished image.
A visual walkthrough of depth cues, changing references, denoising and upscaling in noise painting, with the iterations and self-corrections intact.