Seeking Advice on Krea2 LoRA Training Params
ArchAngelAries · reddit · 2026-07-17
This Reddit discussion asks for advice on tuning parameters for Krea2 LoRA training. The author spent about $20 on Runpod using Ostris's AI Toolkit to train a character LoRA, but the results remain unsatisfactory.
The main issues encountered include:
- Local training difficulties in an AMD Windows environment, forcing a reliance on cloud computing.
- Mediocre results with default parameters; even when halving the learning rate and increasing steps to 3000/4000 for larger datasets, the generated outputs still look artificial.
- With smaller datasets, the model only learns vague concepts, resulting in insufficient character similarity.
- Currently using natural language captions + a unique character trigger word, but the concept remains unstable.
The author is worried that further increasing repeats will "ruin" the model or make it too rigid, and is therefore seeking recommendations for more appropriate steps, rank/alpha, and other configuration tips.
More from Multimodal
- Pablo Stanley shares a full AI video workflow using ChatGPT, Gemini, Runway and CapCut — jdjohnson · 2026-07-21
- Meta AI text input now lets users interleave images with text — ezyang · 2026-07-21
- ShotPlan adds learnable planning tokens for cinematic multi-shot video generation — Tele-AI · 2026-07-21
- Same prompt, Seedance 2 and Grok are compared on cinematic transformation output — LudovicCreator · 2026-07-21
- CG Chefs Showcases Retro Anime Style AI Video Generation — nicolascraske · 2026-07-21
- Night-party video demo uses Seedance 2.0, timecode prompts and 4K upscaling — gen_ericai · 2026-07-21