Krea2 Turbo 2-Step Distill LoRA Hits New Checkpoint, Renders Up to 2048×2048
TimeTruth2490 · reddit · 2026-09-15
Developer lvladikov released a new checkpoint (chk17464) of the Krea 2 Turbo distillation LoRA, cutting generation to roughly a quarter of the base model's steps.
- 2-step fast preview: quality holds up when the subject fills much of the frame (portraits, close-up objects); small subjects — faces in crowds, distant objects — still come out ghosted or smeared, in which case the companion 4-step LoRA is recommended.
- Spend saved steps on resolution: renders up to 2048×2048, Krea's published recommended maximum, are within reach even though the LoRA was trained at 1440×1440. Past 2048², the base model itself duplicates subjects.
- Works with Krea 2 Raw at 7+ steps for some prompts.
The project is still in training, changing one recipe variable at a time and replacing checkpoints only when sweeps and visual checks agree. Full comparisons and known issues are on the Hugging Face model card.
More from Multimodal
- HyperFrames + HeyGen: decent video from a basic prompt in ~10 minutes — toolstelegraph · 2026-10-30
- Hailuo MCP + Claude: chaining motion graphics generation through an agent — aziz4ai · 2026-10-30
- AI short film Nexus ep 7 cost $28,775 to make, creator shares full breakdown — NightsRadiant · 2026-10-30
- LayerShop: open-source Photoshop-style editor with ComfyUI local AI on layers — HonestTicket6013 · 2026-10-30
- AI video prompt goes viral: rally car racing a collapsing volcanic mountainside — umesh_ai · 2026-10-30
- Redditor finishes a short dinosaur movie made with Veo 3 and Seedance — HeadOpen4823 · 2026-10-30