CoolFace
Modelpublic

lodestones/Kroma

sourceHugging Faceotherupdated 7d agoView on Hugging Face
344likes
Model Card

Kroma v0.2 — Krea 2 fine-tune (full model)

Kroma is a fine-tune of Krea 2. As of v0.2 it is distributed as a full model checkpoint — the v0.1 LoRA-delta packaging has been retired. v0.2 is more training on top of v0.1, with no LoRA delta compression, and the Krea 2 Turbo delta merged in as a rank-512 LoRA — a single file gives you the fine-tuned quality at Turbo speed.

  • —File: kroma-v0.2-turbo.safetensors (full diffusion model)
  • —Type: full fine-tuned checkpoint (not a LoRA — no base model required)
  • —License: MIT
Upgrading from v0.1: v0.1 was a LoRA + .diff deltas loaded on top of a Krea 2 checkpoint. v0.2 replaces that entirely — load it as a normal diffusion model and drop the LoraLoader from your workflow.

Running in ComfyUI

1. Requirements

  • —A recent ComfyUI with native Krea 2 support (update to the latest master).
  • —The Krea 2 text encoder and VAE (the diffusion model is this repo — no separate base checkpoint needed).

2. Files to download

Place these in your ComfyUI folders:

WhatWhereNotes
kroma-v0.2-turbo.safetensorsComfyUI/models/diffusion_models/this repo
Krea 2 text encoder (Qwen3-VL)ComfyUI/models/text_encoders/load with CLIPLoader type krea2
Krea 2 VAEComfyUI/models/vae/

3. Sampling settings

The Krea 2 Turbo delta is merged in, so sample with Turbo settings:

SettingValue
Steps8-12
Guidance (CFG)1.0 – 1.5
Shift (mu)1.15
These are the settings the Turbo checkpoint was distilled for; keep them for best results.

4. Workflow

  1. 1.Load Diffusion Model → select kroma-v0.2-turbo.safetensors.
  2. 2.Load CLIP → the Krea 2 text encoder, set type to `krea2`.
  3. 3.Load VAE → the Krea 2 VAE.
  4. 4.KSampler → steps / CFG / shift per the table above.
  5. 5.Prompt and generate.

Minimal node chain:

Load Diffusion Model (kroma-v0.2) ─► KSampler ─► VAE Decode ─► Save Image
Load CLIP (krea2) ─────────────────► (clip)       ▲
Load VAE ───────────────────────────────────────┘
Empty Latent / prompt ────────────────────────────►

5. Tips

  • —If the output looks off, confirm the text encoder is loaded with type `krea2` (Krea 2 expects a 12-layer Qwen3-VL stack) and that you're on a current ComfyUI.

Provenance

v0.2 was produced by continued full fine-tuning of the K2 (Krea 2) checkpoint behind Kroma v0.1 — without the rank-256 LoRA delta compression used for the v0.1 release — followed by merging the Krea 2 Turbo delta in as a rank-512 LoRA.

License

MIT — see LICENSE. Note that Krea 2 is governed by its own license; this fine-tune does not grant any rights to the base weights.