Moss LoRAs Training
Well, after 2 days of training LoRAs, it seems that Moss has trouble learning characters or fine details.
Proof1 community thread
Key points
- It seems that likeness never really converges.
- I've used multiple datasets that I've previously used with Krea 2, Anima, and Z-Image, but it seems that, for now, there's something unusual about Moss that prevents the LoRA from learning any further beyond a certain point.
- Apparently, I'm not the only one experiencing this, as several people have mentioned similar issues on Twitter, Reddit, and Discord.
- The creator of Fizgig has made some progress, but even so, the LoRA still doesn't converge sufficiently.
Sources (1)
- [1]Moss LoRAs Trainingr/StableDiffusion (top, day) · Oct 20, 12:41 AM
Well, after 2 days of training LoRAs, it seems that Moss has trouble learning characters or fine details.
It seems that likeness never really converges.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026People claiming to train a LoRA in 30 mins or less locally...you're lying right?
- Oct 19, 2026Krea 2 Turbo @ 4 steps (w/ my 4-step LoRA) vs MOSS SFT at 4 steps (ww AK's 4-step LoRA) - same prompt set you've seen me publish in the past (headline: MOSS is really good! and great job by AK on his LoRA)
- Oct 19, 2026Are the days of free open-source and shareware over?
- Oct 18, 2026Same prompt, 3 models: SDXL (WAI-Illustrious v17) vs Krea 2 turbo vs MOSS-Image-SFT (anime scenes, no LoRA)
- Oct 17, 2026MOSS-Image LoRA training is now in AcademiaSD LoRAlab Trainer Studio (Base and SFT, under 6 GB VRAM)
- Oct 17, 2026Fizgig Lora Isolation - A new process for more powerful LoRAs
