Kroma v0.2: Krea 2 Full-Model Fine-Tune for ComfyUI

Lodestones released Kroma v0.2 on August 9, and it changes the packaging completely: Kroma is no longer a LoRA. v0.2 is distributed as a full Krea 2 fine-tuned checkpoint, available in both base and Turbo variants on Hugging Face.

What changed from v0.1

Kroma v0.1 shipped as a rank-256 LoRA with fine-tuned norm/modulation weight deltas that had to be loaded on top of a Krea 2 checkpoint. v0.2 retires that packaging:

• Full model, no base required — drop the LoraLoader from your workflow and load kroma-v0.2-turbo.safetensors as a normal diffusion model.
• More training — v0.2 continues fine-tuning beyond v0.1, without the LoRA delta compression.
• Turbo built in — the Krea 2 Turbo delta is merged in as a rank-512 LoRA, so one file gives fine-tuned quality at Turbo speed.
• Two variants: kroma-v0.2-base.safetensors (~51.3 GB) and kroma-v0.2-turbo.safetensors (~25.6 GB).

Community quants are already available: Silveroxides Kroma-Quant ships INT8 convrot variants (~14 GB) and BF16 base-to-turbo conversions, plus a civitai INT8 convrot release for ComfyUI.

How to use Kroma v0.2 in ComfyUI

The fine-tune uses the standard Krea 2 stack, so it works with ComfyUI's native Krea 2 support:

1. Put kroma-v0.2-turbo.safetensors in ComfyUI/models/diffusion_models/
2. Load the Krea 2 text encoder (Qwen3-VL, 12 layers) with a CLIPLoader set to type krea2
3. Load the Krea 2 VAE
4. Sample with the Turbo settings the checkpoint was distilled for:

Download the official Krea 2 Turbo T2I workflow

Availability

Kroma v0.2 is available now on Hugging Face (kroma-v0.2-base.safetensors and kroma-v0.2-turbo.safetensors). The release was announced on Reddit on August 9, with the Silveroxides INT8 quants linked from the thread.