Kroma v0.1: ComfyUI-Ready Krea 2 LoRA With Full Fine-Tune Fidelity

Lodestones released Kroma v0.1, a ComfyUI-ready LoRA fine-tune for Krea 2 that packs a full fine-tuned model into a single safetensors file.

What Makes Kroma Different

Most LoRAs only store low-rank adapter weights, which is why they drift from the original fine-tune. Kroma takes a different approach: it was rank-reduced down to rank 256 / alpha 256 and also carries the fully fine-tuned normalization and modulation tensors as weight deltas (.diff). Loading it on top of the base model reproduces the original fine-tuned behavior instead of an approximation.

• File: kroma-v0.1.safetensors (~1.88 GB)
• Contents: 264 LoRA adapters plus 159 RMSNorm / modulation weight deltas
• Compatibility: works with both Krea 2 checkpoints

Base vs Turbo

Kroma works on both Krea 2 base and Krea 2 Turbo. Turbo is recommended, since the checkpoint was distilled for its sampling settings:

How to Use Kroma in ComfyUI

Kroma is packaged as a standard ComfyUI LoRA, so no custom nodes are required. The repo's README notes the .diff deltas are applied automatically by ComfyUI's LoraLoader.

1. Put kroma-v0.1.safetensors in ComfyUI/models/loras/
2. Load Krea 2 (base or Turbo) in a Load Diffusion Model node
3. Load the Krea 2 text encoder with a CLIPLoader set to type krea2
4. Load the Krea 2 VAE
5. Add a LoraLoader (Model only) between the model loader and the sampler, select kroma-v0.1.safetensors at strength 1.0 (lower to 0.7–0.9 for a subtler effect)
6. Set the sampler per the table above, then prompt and generate

Download the official Krea 2 Turbo T2I workflow