FLUX.1 [dev] VRAM Requirements
FLUX.1 [dev] needs about 12.1 GB of VRAM with FP8 weights at 1024×1024, loading one model at a time. In FP16 it needs 23.2 GB (35.4 GB with everything loaded), and GGUF Q4 brings it to 7.3 GB. The smallest consumer GPU that runs it is the RTX 3050 8GB with Q4 weights, or the RTX 4060 Ti 16GB in FP8.
12.1 GB peak VRAM
FLUX.1 [dev] at 1024×1024 with FP8 weights, loading one model at a time (ComfyUI default on smaller cards). Keeping everything loaded needs 19.7 GB.
- Diffusion model 11.1 GB
- Text encoders 4.5 GB
- VAE 0.2 GB
- Peak activations 3.1 GB
- Overhead 0.8 GB
GPU fit
- RTX 4060 8GB62% on GPU
- RTX 3060 12GB93% on GPU
- RTX 4060 Ti 16GBFitswith swapping
- RTX 4090 24GBFitsall loaded
- RTX 5090 32GBFitsall loaded
- RTX 6000 Ada 48GBFitsall loaded
- RTX PRO 6000 Blackwell 96GBFitsall loaded
- Apple M4 Max 128GBFitsall loaded
The VAE decode is the peak here. Turn on tiled VAE decoding to cut it to about 0.8 GB.
FLUX.1 [dev] VRAM by precision
At 1024×1024. Text encoders in FP8 except for the FP16 row.
| Weights | Diffusion model | Peak (swapping) | All loaded | Smallest tier |
|---|---|---|---|---|
| FP16 / BF16 | 22.2 GB | 23.2 GB | 35.4 GB | 32 GB |
| FP8 | 11.1 GB | 12.1 GB | 19.7 GB | 16 GB |
| GGUF Q8_0 | 11.8 GB | 12.8 GB | 20.4 GB | 16 GB |
| GGUF Q5_K_S | 7.6 GB | 8.6 GB | 16.3 GB | 10 GB |
| GGUF Q4_K_S / NF4 | 6.3 GB | 7.3 GB | 15.0 GB | 8 GB |
VRAM by resolution
FP8 weights, with model swapping.
| Resolution | Latent tokens | Activations | Peak VRAM |
|---|---|---|---|
| 1024×1024 | 4,096 | 0.2 GB | 12.1 GB |
| 1536×1536 | 9,216 | 0.4 GB | 12.3 GB |
About FLUX.1 [dev]
A 12B guidance-distilled model, so it runs one forward pass per step instead of two, which halves activation memory compared with classic CFG. The FP16 weights are about 22 GB, so most people run FP8 or GGUF Q8_0 (about 12 GB) or Q4 (about 7 GB).
| Component | Size |
|---|---|
| Diffusion model | 11.9B parameters (Rectified-flow transformer) |
| Text encoders | CLIP ViT-L (0.123B), T5-XXL encoder (4.76B) |
| VAE | 0.084B parameters |
| Classifier-free guidance | Distilled (1 pass per step) |
| Released | 2024-08 |
| License | FLUX.1 [dev] Non-Commercial License |
FLUX.1 [dev] VRAM: common questions
Can FLUX.1 dev run on 12GB VRAM?
Yes with GGUF Q4 to Q5 or NF4 weights and T5 in FP8 or offloaded. FP8 weights on 12 GB rely on model swapping and partial offload, which is slower.
Is FLUX.1 dev FP8 much worse than FP16?
The difference is small for most prompts. GGUF Q8_0 is closer to FP16 than plain FP8 at a similar size.
How much VRAM does FLUX.1 [dev] need?
At its default 1024×1024: about 23.2 GB in FP16 with model swapping (35.4 GB with everything loaded), 12.1 GB in FP8 and 7.3 GB with GGUF Q4 weights.