Krea 2 skin-texture LoRA comes with dataset, training guide, and settings
ashishsanu · reddit · 2026-07-27
The author publishes a skin-texture LoRA trained for Krea 2, along with the dataset and a training guide.
Key details:
- Best results were reportedly obtained on Krea 2 Turbo with 28 steps, guidance 4.5, and Euler/Simple sampling.
- Recommended strength is 0.6–1.0, with lower values for blending.
- The LoRA aims to produce more realistic skin detail, but has drawbacks: it tends to skew younger prompts older, and framing drifts toward head-and-shoulders.
- Training was done with PEFT LoRA, rank 16/alpha 16, 1,500 steps, batch size 1, on 26 images.
- The author also explains hardware requirements and notes that 512px training can fit on 16GB, while 1024px needs roughly 32GB+.
The dataset, model, and guide are all linked publicly on Hugging Face and Civitai.
Related event: Krea 2 Skin Texture LoRA Dataset and Guide Released(2 posts)→
More from Multimodal
- Screenshot any character, one prompt recreates it — iamaliveix · 2026-09-23
- Third-party test: Claude Opus 5.5 renders finer 3D scenes but costs 13x more than GPT-6 Sol — testingcatalog · 2026-09-23
- ComfyUI trick: aux preprocessor + Qwen transfers poses across characters with one prompt — Acceptable-Work8202 · 2026-09-23
- Same portrait prompt across Midjourney V6.1, V7 and V8.2: do older models look better? — tisch_eins · 2026-09-23
- Testing AI character consistency across a 20-image travel sequence — SiennaVaire · 2026-09-23
- Midjourney v8.2 Faces: New Portrait Generation Samples Shared — azed_ai · 2026-09-23