Redditor Shares Qwen 2.1 Character LoKr Training Recipe: Multi-Resolution Is Key
Any_Tea_3499 · reddit · 2026-09-24
A Redditor detailed their first Qwen 2.1 character LoKr runs in AI Toolkit: LoKr Rank 4, lr 0.0001, weight decay 0.0002, Sigmoid, batch size 1, and training across 512/768/1024 resolutions (1024 alone caused issues), with datasets of just 30-70 images. For generation they stack a Lenovo lora at 0.6-0.7 strength for realism, plus the Viggle Qwen 2.1 turbo lora at 0.5 strength, 15 steps, 2.5 CFG with Resmultistep and Beta. They found Qwen 2.1 knows far more facial expressions than Krea, yielding more accurate subject likeness.
More from Multimodal
- MiniMax H3-Image image model debuts on Tencent's Miora, free to try Nov 11–17 — aziz4ai · 2026-11-11
- MiniMax H3 hits 100 days open-source: 1,472 derivative models, 148M+ downloads — linoy_tsaban · 2026-11-11
- MiniMax H3-Image explodes an espresso machine photo into 15 labeled parts — Charlietest123 · 2026-11-11
- Indie dev built a tool that clips best moments from 4-hour streams in 30 minutes — kissaadams · 2026-11-11
- Non-Blender user makes a video in 30 minutes with Claude and Seedance — FinanceYF5 · 2026-11-11
- Whimsical Woodland: A Watercolor AI Art Prompt Template for Dreamy Forest Scenes — LudovicCreator · 2026-11-11