MOSS Character LoRas
At this point I’ve trained 2 character Lora’s using datasets that worked good on other models such as Krea 2 and Qwen 2.1.
Proof1 community thread
Key points
- However with MOSS the identity is only about 70-80% there.
- Have others experienced the same issues?
- I used AI toolkit both the base and sft model.
Sources (1)
- [1]MOSS Character LoRasr/StableDiffusion (top, day) · Oct 19, 01:26 AM
At this point I’ve trained 2 character Lora’s using datasets that worked good on other models such as Krea 2 and Qwen 2.1.
However with MOSS the identity is only about 70-80% there.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026A comparison of 6 open source models. Krea2, Qwen2.1, Flux3 Klein, Flux 2 Klein, ZIT, & Moss. Same prompt used for all.
- Oct 18, 2026T2I Comparison (~4MP): Qwen Image 2.1 Turbo vs. MOSS-Image-SFT vs. Krea 2 Turbo vs. Flux 3 Klein (🤮)
- Oct 17, 2026Nemotron 3.5 Bolt, Muse Spark Open, and a Fast LFM2.5 24B
- Oct 16, 2026FLUX 3 vs KREA 2 vs Z image turbo vs Qwen 2.1
- Oct 16, 2026China tightens export controls on open-weight models and 23 key AI technologies
- Oct 15, 2026Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint