Krea 2
A 12.8B single-stream transformer trained from scratch by Krea, released as Raw (undistilled, for training) and Turbo (8 steps). It shares the Qwen3-VL 4B text encoder with FLUX.3 Image Klein and uses Qwen-Image's VAE. Commercial use is capped at US$1M annual revenue.
What you actually download
A diffusion pipeline is three separate models. These sizes are read from the real .safetensors files in the Hugging Face repository, not estimated.
Peak VRAM at 1024×1024, one image
Every combination of precision and offload strategy. Lower precision costs image quality; offloading costs speed.
| Precision | Everything in VRAM | Text encoder on CPU | Sequential CPU offload |
|---|---|---|---|
| BF16 / FP16Reference quality. What the repos ship by default. | 38.2 GB | 29.3 GB | 29.1 GB |
| FP8Roughly half the weight memory. Small quality cost, widely used for Flux and SD 3.5. | 20.6 GB | 16.2 GB | 15.9 GB |
| NF4 / INT4Aggressive 4-bit. Visible quality loss, but it puts big models on small cards. | 15.4 GB | 10.9 GB | 10.7 GB |
Which machines can run it
62 of the 79 devices we profile can run Krea 2 at 1024×1024 in some configuration. Showing the best-quality setup each one supports.
Showing the 30 devices with the most memory. The full hardware matrix covers all 79.
Licence: Krea 2 Community License
With conditions. Krea 2 is released under the Krea 2 Community License (v1, 22 June 2026). Non-commercial use is open. Commercial use of the model, derivatives or the images it generates is permitted only if you and your affiliates have under US$1M in trailing twelve-month revenue; above that you need an enterprise licence from Krea. Everyone must run content filters, and redistributed derivatives must be named starting with "Krea".
Compare model licences →Krea 2: common questions
- How much VRAM does Krea 2 need?
- At 1024x1024 and batch 1, Krea 2 peaks at about 38.2GB with everything resident at BF16, and as little as 10.7GB at NF4 + sequential CPU offload. The weights alone are 35.4GB at BF16 — 26.3GB of transformer, 8.9GB of text encoder and 0.25GB of VAE.
- Which GPUs can run Krea 2?
- 62 of the 79 devices we profile can run it at 1024x1024 in some configuration. The Apple Mac Studio (M5 Ultra, 512GB, 2026) handles it best, at about 38.2GB. Lower precision and text-encoder offload widen that list considerably.
- Can Krea 2 be used commercially?
- With conditions. Krea 2 is released under the Krea 2 Community License (v1, 22 June 2026). Non-commercial use is open. Commercial use of the model, derivatives or the images it generates is permitted only if you and your affiliates have under US$1M in trailing twelve-month revenue; above that you need an enterprise licence from Krea. Everyone must run content filters, and redistributed derivatives must be named starting with "Krea".
- Does offloading the text encoder help on Krea 2?
- Yes, substantially. Krea 2 uses a large language model as its text encoder — 8.9GB at full precision. Encoding happens once per prompt, so moving it to CPU costs a couple of seconds and frees that 8.9GB for the rest of the run, taking peak VRAM at BF16 from 38.2GB down to 29.3GB.
Weight sizes read from the model repository on 2026-09-12. Activation memory is an approximation, not a benchmark — BFWR does not run a hardware test lab.
Other open image models
Guides that go deeper
Setting this model up in ComfyUI, its licence, and the hardware it needs.
- Krea 2 Locally: Open Weights, VRAM and What the Licence AllowsTurbo versus Raw, the 18.6GB template download, and the US$1M cap on commercial use.Read the guide
- Qwen-Image 2.1 in ComfyUI: Files, VRAM, Turbo and the LicenceWhat to download, the memory it needs, and its research-only licence.Read the guide
- How to Run Stable Diffusion Locally: Hardware, VRAM and SetupCurrent models from SD 1.5 to Qwen-Image, and the VRAM each needs.Read the guide