FLUX.3 Image [klein]
The first open FLUX.3 model: 5.6B parameters, text-to-image and editing with up to 10 references in one checkpoint, and a Qwen3-VL 4B text encoder. Black Forest Labs says not to use it without its 9B prompt upsampler. Free commercial use of outputs applies only below US$5M annual revenue. Black Forest Labs measured a 31GB peak at BF16 on a B200 with its reference code, higher than the estimate on this page; treat these figures as a floor.
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. | 23.1 GB | 14.2 GB | 13.9 GB |
| FP8Roughly half the weight memory. Small quality cost, widely used for Flux and SD 3.5. | 13.1 GB | 8.7 GB | 8.4 GB |
| NF4 / INT4Aggressive 4-bit. Visible quality loss, but it puts big models on small cards. | 10.9 GB | 6.5 GB | 7.2 GB |
Which machines can run it
75 of the 79 devices we profile can run FLUX.3 Image [klein] 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: FLUX Kommunity License
With conditions. FLUX.3 Image [klein] is released under the FLUX Kommunity License v1.0. Use is free for non-commercial purposes. If you and your affiliates have under US$5M in gross annualized revenue, you may also use the images it generates commercially, provided you filter or review outputs and disclose AI generation where the law requires it. Commercial fine-tunes, serving the model to others and larger companies need a licence from Black Forest Labs.
Compare model licences →FLUX.3 Image [klein]: common questions
- How much VRAM does FLUX.3 Image [klein] need?
- At 1024x1024 and batch 1, FLUX.3 Image [klein] peaks at about 23.1GB with everything resident at BF16, and as little as 6.5GB at NF4 + text encoder on CPU. The weights alone are 20.3GB at BF16 — 11.1GB of transformer, 8.9GB of text encoder and 0.34GB of VAE.
- Which GPUs can run FLUX.3 Image [klein]?
- 75 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 23.1GB. Lower precision and text-encoder offload widen that list considerably.
- Can FLUX.3 Image [klein] be used commercially?
- With conditions. FLUX.3 Image [klein] is released under the FLUX Kommunity License v1.0. Use is free for non-commercial purposes. If you and your affiliates have under US$5M in gross annualized revenue, you may also use the images it generates commercially, provided you filter or review outputs and disclose AI generation where the law requires it. Commercial fine-tunes, serving the model to others and larger companies need a licence from Black Forest Labs.
- Does offloading the text encoder help on FLUX.3 Image [klein]?
- Yes, substantially. FLUX.3 Image [klein] 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 23.1GB down to 14.2GB.
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.
- FLUX.3 Image Klein in ComfyUI: Files, VRAM and the $5M Licence RuleThe prompt upsampler, the 17.7GB of files and the revenue limit on commercial use.Read the guide
- Flux LoRA: Which Base Model You Can Legally Train OnA LoRA is a Derivative; only klein 4B and schnell are Apache-2.0.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