Omnico/Chroma1_diff_loras

Extracted Loras from Chroma’s checkpoints

<Gallery />

Model description

The LoRAs were extracted from fine-tuned checkpoints and official releases, with chroma-unlocked-v46-detail-calibrated also used as a reference for comparison. I did this primarily for myself: LoRAs are simply easier to work with — you can mix them in real time and set different weights and strengths. They also save storage space on your local machine. Choose a rank to suit your preferences and hardware capabilities (the lower the rank, the less memory is required).

After extraction, the weights were normalized. There is a direct correlation between rank and the acceptable LoRA strength range: for r64, keep the strength up to 130% at most; for r384, up to roughly 250%.

As we know, Chroma1 is a highly creative, uncensored model, but it is extremely difficult to work with. The main issues I see are unpredictable results, slow speed (even in Turbo Flash mode), and noise. This LoRA extraction project was an attempt to simplify setup and generation on Chroma. I did not fully achieve that goal — it is still a challenging model to work with. That said, if you find the extracted LoRAs useful and get good results, please share them with the community. Chroma1 is an underrated, capricious, yet remarkably powerful model.

A lot of time, electricity, and compute went into this on my small and not particularly powerful Mac Mini M4. But it was an interesting research project.

Download model

Download them in the Files & versions tab.