FLUX.1 schnell VRAM requirements
Short answer: FLUX.1 schnell's files total 33.7 GB, and the transformer alone is 23.8 GB in BF16.[1] Hugging Face Diffusers says Flux needs ~50 GB of RAM/VRAM to load every component, or under 16 GB with an FP8 transformer and T5 plus CPU offload.[2]
FLUX.1 schnell is Black Forest Labs' fast, few-step text-to-image model. It is licensed Apache 2.0, and Hugging Face lists 11,891,178,560 parameters for its single-file transformer.[1] It is the same size as FLUX.1 dev, which is not Apache-licensed.
Weights only: file sizes on Hugging Face
Exact sizes from the black-forest-labs/FLUX.1-schnell file listing (1 GB = 1,000,000,000 bytes; split files added; checked November 4, 2026).[1]
| Component | Files | Size |
|---|---|---|
| Transformer (BF16) | transformer/ (3 files) or flux1-schnell.safetensors |
23.78 GB |
| T5-XXL text encoder | text_encoder_2/ (2 files) |
9.52 GB |
| CLIP text encoder | text_encoder/ |
0.25 GB |
| VAE | vae/ |
0.17 GB |
| Total (weights only) | 33.72 GB |
Memory while generating (official figures only)
Image models also need activation memory, which grows with resolution and batch size. Black Forest Labs publishes no VRAM figure. The only official numbers we found are from the Diffusers Flux docs, which cover both FLUX.1 models:[2]
| Reference setup (Diffusers docs) | What the docs say |
|---|---|
| All components loaded at full precision | "~50GB of RAM/VRAM" |
| FP8 transformer + FP8 T5 (optimum-quanto) + model CPU offload | "less than 16GB of VRAM" |
| Sequential CPU offload + VAE slicing/tiling | For "low vram GPUs (i.e. between 4 and 32 GB VRAM)"; slow |
We don't publish our own activation estimate, because we haven't measured one. See the FLUX VRAM guide for every low-VRAM option.
What this means for your GPU
- 24 GB (RTX 3090/4090): the BF16 transformer alone is 23.8 GB, so you need offload or FP8. See what you can run on an RTX 4090.
- 16 GB and under: use the FP8 + offload setup above.
- No GPU? The cheapest GPU rental per hour lists 48 GB and 80 GB cards.
Frequently asked questions
How much VRAM does FLUX.1 schnell need?
About 50 GB to load every component at full precision, per the Diffusers docs, or under 16 GB with FP8 and CPU offload. The files themselves total 33.7 GB.
Can FLUX.1 schnell run on 12 GB or 16 GB?
Yes with FP8 and CPU offload (under 16 GB per Diffusers), or sequential offload on 4–32 GB cards, which is slower.
Is FLUX.1 schnell smaller than FLUX.1 dev?
No. Both have the same 23.8 GB BF16 transformer. Schnell is distilled for fewer steps and is Apache 2.0 licensed.
Every number on this page links to its source.
How we calculate + sources ⓘ
Sources
All sources checked on .
- Hugging Face, black-forest-labs/FLUX.1-schnell (file listing, parameter count, license; checked November 4, 2026). https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main
- Hugging Face Diffusers, Flux pipeline docs, "Optimize" (checked November 4, 2026). https://huggingface.co/docs/diffusers/api/pipelines/flux#optimize