RawSnap v2
retained text
v2 of my RawSnap LoRA for Krea 2 is up on Civitai. Same idea as v1: photos that look taken on a phone, not rendered — real skin texture, harsh flash, grain, awkward framing. What changed: 3300 more training steps, merged with v1 into a single file. No trigger word. Strength works between 1.0 and 1.5, sweet spot around 1.2. It can be more noisy than v1 — honestly, that's part of the look. Settings I use: 8 steps, CFG 1.0, Euler, on Krea 2 Turbo. Trained locally on an RTX 5060 (8GB). The comparisons attached are v1 vs v2, same prompt and seed. Civitai: https://civitai.com/models/2999833/rawsnap-krea2 Feedback is welcome, especially from anyone stacking it with a character LoRA.
document fetched 2026-10-25 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1xe8siy/rawsnap_v2