>>106511901
You want to train at least 2,000 steps and can train one (latent) image up to 100 times on diffusion models. 20x100=2,000
Also more is not always better (with smol brain SDXL-based models), but Chroma can take as many good reference images as you throw at it