run-diffusion-xl

For best outputs:

• Sampler: Restart (realistic pictures) / DPM++ SDE (the rest)

• Steps: 20, cfg 7, CLIP skip 2

• VAE: sdxl_vae.

• Simple negs: cartoon, anime, painting, oil painting, deformed, ugly, bad quality, low quality, nsfw, nude, black and white, bw, b&w,

• Tip: add "desaturated, 35mm photo, vintage, stained, highly detailed, found footage, Holga 120n, film grain, vignette, (intricate details), (hyperdetailed), 8k hdr, high detailed, lot of details, high quality, soft cinematic light, dramatic atmosphere, atmospheric perspective" at the end of your pos to increase realism.

• Important: The model tends to favor closeups or "headshots" more than full body unless you explicitly prompt for that.