Check my GPU →

FLUX.1 Fill [dev] on 20 GB of VRAM

Yes. Q8_0 fits on 20 GB with 4.7 GB to spare. The inpainting and outpainting model from the FLUX.1 Tools release. Same 12B size class as FLUX.1 dev.

Image model · 12BAny 20 GB GPUData 2026-10-01
Runs well
Yes.

Take Q8_0 (12.7 GB). In use it comes to about 15.3 GB, with 4.7 GB left over on 20 GB.

Best fileQ8_0
File size12.7 GB
VRAM needed~15.3 GB
System RAM32 GB+
Memory map · 20 GB GPU15.3 GB / 20 GB
010 GB20 GB
Weights Q8_0 · 12.7 GBWorking memory · 1.8 GBReserve · 0.8 GBFree · 4.7 GB

01What to download

PartFileFolderSize
Modelflux1-fill-dev-Q8_0.gguf
Q8_0 · YarvixPA/FLUX.1-Fill-dev-GGUF
models/unet12.7 GBDownload →
Text encodert5xxl_fp8_e4m3fn.safetensors
T5-XXL FP8 · comfyanonymous/flux_text_encoders
models/text_encoders4.9 GBDownload →
Text encoderclip_l.safetensors
CLIP-L · comfyanonymous/flux_text_encoders
models/text_encoders0.2 GBDownload →
VAEae.safetensors
FLUX.1 VAE (ae) · Comfy-Org/Lumina_Image_2.0_Repackaged
models/vae0.3 GBDownload →
Total download · keep about the same free on disk18.2 GB

Alternatives: T5-XXL FP16 (9.8 GB). Sizes from Hugging Face, 2026-10-01.

02Every file, on 20 GB

FileSizeNeededOn this cardQualityDownload
16-bit
SAFETENSORS · black-forest-labs
23.8 GB26.4 GB6.4 GB too bigthe original weightsHugging Face →
Q8_0 ←
GGUF · YarvixPA
12.7 GB15.3 GBfits · 4.7 GB sparepractically identical to the originalHugging Face →
Q6_K
GGUF · YarvixPA
9.9 GB12.5 GBfits · 7.5 GB sparevery close to the originalHugging Face →
Q5_K_S
GGUF · YarvixPA
8.3 GB10.9 GBfits · 9.1 GB spareclose; small differences in fine detailHugging Face →
Q4_K_S
GGUF · YarvixPA
6.8 GB9.4 GBfits · 10.6 GB sparegood; some loss in fine detail and textHugging Face →
Q3_K_S
GGUF · YarvixPA
5.2 GB7.8 GBfits · 12.2 GB sparenoticeable loss of detailHugging Face →

“Needed” adds 1.8 GB of working memory for FLUX.1 Fill and a 0.8 GB reserve. Sizes 2026-10-01.