1 step img2img with a latent consistency model is wild. The quality and speed of the outputs is amazing. 1,3: Input image
2,4: img2img with 1 step (0.65 and 0.5 prompt strengths)

→ View original post on X — @fofrai