RawSnap v3 is here — LoKr

What's new in v3 :

Retrained from scratch as a LoKr (not a standard LoRA) on an expanded dataset of 749 images (11,235 steps, 15 epochs)

⭐ The recommended setup

Best results — stack them:

RawSnap v3 (epoch 15 or epoch 12) 1.2

RawSnap v2 0.35

RawSnap [Krea2] — v2

3300 more training steps and merged with v1 into one file.

Strength: works between 1.0 and 1.5 — sweet spot 1.2.

It may be more noisy than v1.

Real phone-photo energy, zero AI gloss. This LoRA pushes Krea 2 Turbo toward

candid, true-to-life smartphone photography — harsh direct flash, night shots,

visible sensor grain, awkward framing, and the beautiful mess of real life.

What it does:

- Natural skin with real texture and pores — no plastic smoothing

- Harsh direct flash blowout, night shots, motivated indoor/outdoor light

- Visible phone sensor noise and grain

- Awkward, random, tilted framing — like a real snapshot

- Unposed, spontaneous-feeling people and expressions

- Lived-in rooms, streets, and everyday settings

Trigger word:

There is NO trigger word. Just describe your scene naturally.

Do NOT add "photorealistic", "shot on iPhone", "masterpiece", "8k", or

"ultra detailed" — those make results stiffer and fight the LoRA.

LoRA strength: Recommended between 1.5 - 2.0