FLUX.2 [dev]
The text encoder is the story: Black Forest Labs couples a Mistral-3 24B vision-language model to the transformer, so the encoder alone is 48GB at BF16 - larger than the whole of FLUX.1. Its own repository says it needs an H100-equivalent GPU.
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. | 115.6 GB | 67.6 GB | 67.3 GB |
| FP8Roughly half the weight memory. Small quality cost, widely used for Flux and SD 3.5. | 59.4 GB | 35.4 GB | 35.0 GB |
| NF4 / INT4Aggressive 4-bit. Visible quality loss, but it puts big models on small cards. | 46.5 GB | 22.5 GB | 26.8 GB |
Which machines can run it
22 of the 75 devices we profile can run FLUX.2 [dev] 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 75.
Licence: FLUX [dev] Non-Commercial License v2.0
No. FLUX.2 [dev] is released under FLUX [dev] Non-Commercial License v2.0, which does not permit commercial use. You would need a separate agreement with the rights holder to ship output commercially.
Compare model licences →FLUX.2 [dev]: common questions
- How much VRAM does FLUX.2 [dev] need?
- At 1024x1024 and batch 1, FLUX.2 [dev] peaks at about 115.6GB with everything resident at BF16, and as little as 22.5GB at NF4 + text encoder on CPU. The weights alone are 112.8GB at BF16 — 64.5GB of transformer, 48.0GB of text encoder and 0.34GB of VAE.
- Which GPUs can run FLUX.2 [dev]?
- 22 of the 75 devices we profile can run it at 1024x1024 in some configuration. The Apple Mac Studio (M5 Ultra, 512GB, 2026) handles it best, at about 115.6GB. Lower precision and text-encoder offload widen that list considerably.
- Can FLUX.2 [dev] be used commercially?
- No. FLUX.2 [dev] is released under FLUX [dev] Non-Commercial License v2.0, which does not permit commercial use. You would need a separate agreement with the rights holder to ship output commercially.
- Does offloading the text encoder help on FLUX.2 [dev]?
- Yes, substantially. FLUX.2 [dev] uses a large language model as its text encoder — 48.0GB at full precision. Encoding happens once per prompt, so moving it to CPU costs a couple of seconds and frees that 48.0GB for the rest of the run, taking peak VRAM at BF16 from 115.6GB down to 67.6GB.
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
Which GPUs have the VRAM these models need, and what they cost right now.
- 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
- 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 Increase VRAM: What Actually Works on PC, Laptop and MacIt depends where the memory physically lives.Read the guide