Krea 2 Open Weights: Raw (Undistilled) vs Turbo (Distilled) Image Models
Melvin Vivas · X video post · 2026-06-23 · 0:24 · 159 views · Open on X
Topics: Fine-tuning & Model Customization, Industry Trends & Job Market · Level: intermediate
Summary
This is a short post sharing Krea's announcement that it has released the open weights of its Krea 2 image generation model. There are two versions. Krea 2 Raw is an undistilled checkpoint from mid-training, meant to be fine-tuned. Krea 2 Turbo is a faster distilled version with a wide range of aesthetic styles. The post shows the usual split between a base model you customize and a distilled model you run for fast inference.
Key points
- Krea has released the open weights of Krea 2, a text-to-image generation model.
- Krea 2 Raw is an undistilled checkpoint taken from mid-training. It is meant as a base for fine-tuning.
- Krea 2 Turbo is a distilled version built for fast generation, with a wide range of aesthetic styles.
- Undistilled base models are usually better for fine-tuning, for example with LoRA. Distilled models trade some flexibility for fewer inference steps and lower latency.
- Open weights let you run, customize and self-host the model instead of depending on a closed API.
Resources mentioned
- Krea 2 Raw · tool · huggingface.co · free
An undistilled open-weights image generation checkpoint from mid-training, meant to be fine-tuned. - Krea 2 Turbo · tool · huggingface.co · free
A fast, distilled open-weights version of Krea 2 with a wide range of aesthetic styles. - Krea · tool · krea.ai · check price
An AI image and video generation platform; the company behind the Krea 2 models.
Try this
- Read the details in Krea's announcement thread about the Krea 2 release.
More in Fine-tuning & Model Customization
- Small Local Models + Fine-Tuning Instead of More Compute (Liquid AI, LEAP)
- Evolution of Image LoRAs: Stable Diffusion → Flux → Krea 2
- Open-Source JSONL Viewer for Fine-Tuning Datasets
- Paper: Context-Aware RL for Agentic and Multimodal LLMs
- 1-bit Bonsai Image 4B: A Diffusion Model Under 1GB
- Unsloth joins the PyTorch Ecosystem