Krea2 retro anime LoRA is trained on 18,000 cel-animation frames
Honest_Concert_6473 · reddit · 2026-07-27
A new Krea2 RetroAnime Style LoRA was trained on 18,000 carefully selected cel-animation screenshots to recreate a 1990s–early 2000s retro anime look.
- The author says Krea2 already produces accurate cel-animation style, but the LoRA is meant for users who want a more specific retro-anime flavor.
- It can be used at lower weights to blend with Krea2’s default anime style.
- The model was trained with booru-style tag captions, no trigger words, and is intended to work by simply applying the LoRA.
- Training details are shared: RTX A6000 48GB, 1024px resolution, rank 64, AdamW, LR 1e-4, batch size 4, gradient accumulation 4, constant scheduler, about 6 epochs / 6,000 steps.
- The author notes an interesting side effect: the LoRA sometimes changes poses, clothing and composition, not just style.
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