Chroma-Flash-Heun low-step LoRa

LORA
Reprint


Updated:

30-50 steps are recommended for base Chroma. These LoRas reduce that number depending on the version.

In my opinion, Rank 64 allows high-quality generations with 15-25 steps, great for the free tier on Tensor.

Rank 128 is good for 10 steps. costing only 0.4 credits, but cfg 1 means negative prompts are ignored. Prompt adherence and anatomy are said to be minimally worse. Probably worth it for 125 images per day on the free tier.

If anyone wants extra versions of the LoRa, just comment and I’ll upload some more.

COPIED FROM CIVITAI:

This series of LoRA models are extractions from chroma-unlocked-v47-flash-heun and despite the indicating name it is not restricted to be used with Heun sampler nor that lowest step is supported across all ranks.

I can't upload every model in one go so for any of the other ranks in the meantime, visit this huggingface model repo

Version Detail

Chroma
Rank 128 should be used without CFG (set to 1.0 in ComfyUI) and has significant reduction of step count depending on sampler/scheduler used. (8-16 for dpmpp_sde/heun/res_2s and 16-24 for euler/er_sde/res_multistep/res_2m etc.)

Project Permissions

Model reprinted from : https://civitai.com/models/2032955

Reprinted models are for communication and learning purposes only, not for commercial use. Original authors can contact us to transfer the models through our Discord channel --- #claim-models.

Related Posts