New Krea 2 Turbo Distillation LoRA: 4-step generation with 13% reduced error gap
TimeTruth2490 · reddit · 2026-08-22
The author released a new 4-step distillation LoRA for Krea 2 Turbo, reducing the minimum usable step count from 8 to 4. The new checkpoint (chk00006000) uses an int8 teacher instead of NF4, achieving a 13% smaller held-out gap compared to the full 8-step teacher model. The project emphasizes a loop of train-assess-adapt, releasing checkpoints only when they are measurably better on a held-out set, rather than just training longer. The latest file acts as a rolling pointer to the newest weights for seamless updates.
More from Multimodal
- Atlas demos turn a single image into an explorable 3D scene, generating off-screen areas live — XRarchitect · 2026-10-16
- HyperFrames launches Reference Any Video and auto-applied Brand Kits for AI video — aftahi_ai · 2026-10-16
- Krea 2 + LoRA still beats Flux 3 Klein in pixel-art style test — Angaisb_ · 2026-10-16
- World Labs starts rolling out Atlas beta with pixel-perfect camera path control — theworldlabs · 2026-10-16
- Krea 3 image editing model lands on fal: high-quality edits in under 6 seconds — gorkem · 2026-10-16
- Tripo P2 vs Meshy 7.1: benchmarking today's best image-to-3D models for game assets — noahsolomon · 2026-10-16