← sources

I made ZenTrainer: an open-source LoRA trainer that runs inside ComfyUI (image + video models, live previews, NVIDIA or AMD)

source: reddit (reddit:StableDiffusion)published 2026-10-30 16:18ingested 2026-10-31 02:38

retained text

Hi all, I've been building ZenTrainer, a LoRA trainer that lives inside ComfyUI, and 0.4.0 is out today. It's free and open source (MIT).

Trailer: https://www.youtube.com/watch?v=YS4BkGonflg
GitHub: https://github.com/Nynxz/ZenTrainer

**What it does**

- Trains LoRAs from a ComfyUI tab: pick a model, drop in your pictures, and it fills in the files, captions and settings
- Live previews as it trains, and for characters a likeness score per checkpoint, so you can see which one is best instead of guessing
- Runs on NVIDIA and AMD (ROCm, Linux and Windows)
- Uses the models you already have in ComfyUI, wherever they are

**Models trained end to end**

MOSS Image, Krea 2, Fibo 2.0, FLUX 3 Image, FLUX.2 Klein 9B, Qwen Image 2.1, Z-Image, Anima, SDXL (and its finetunes: Pony, Illustrious and the rest train as they are) and MiniMax H3 for video. Anything else ComfyUI knows shows up as untested.

**Install**

ComfyUI Manager → search "ZenTrainer" → restart → click the lotus in the top bar. Or run it with its own pinned ComfyUI, so a ComfyUI update can never break a run.

The two pixel-art LoRAs in the trailer are community ones, shown with their creators' permission: 4Bit Alchemy by Addictedd and Nostapixx by JadedScorpion.

Happy to answer questions, and bug reports are very welcome

document fetched 2026-10-31 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xjjst7/i_made_zentrainer_an_opensource_lora_trainer_that