Check my GPU →

ERNIE-Image (Turbo) on 32 GB of VRAM

Yes. 16-bit fits on 32 GB with 13.6 GB to spare. Baidu's 8B model from April 2026, strong at rendering text in images. Turbo runs in about 8 steps; Base takes 50.

Image model · 8.03BAny 32 GB GPUData 2026-10-01
Runs well
Yes, with room to spare.

16-bit (16.1 GB) needs about 18.4 GB in use, so 13.6 GB of the 32 GB stays free: room for a larger image, a LoRA or a ControlNet.

Best file16-bit
File size16.1 GB
VRAM needed~18.4 GB
System RAM32 GB+
Memory map · 32 GB GPU18.4 GB / 32 GB
016 GB32 GB
Weights 16-bit · 16.1 GBWorking memory · 1.5 GBReserve · 0.8 GBFree · 13.6 GB

01What to download

PartFileFolderSize
Modelernie-image-turbo.safetensors
16-bit · Comfy-Org/ERNIE-Image
models/diffusion_models16.1 GBDownload →
Text encoderministral-3-3b.safetensors
Ministral 3 3B · Comfy-Org/ERNIE-Image
models/text_encoders7.7 GBDownload →
VAEflux2-vae.safetensors
FLUX.2 VAE · Comfy-Org/ERNIE-Image
models/vae0.3 GBDownload →
Total download · keep about the same free on disk24.1 GB

Only for some workflows: ERNIE prompt enhancer (optional) (6.9 GB). Sizes from Hugging Face, 2026-10-01.

02Every file, on 32 GB

FileSizeNeededOn this cardQualityDownload
16-bit ←
SAFETENSORS · Comfy-Org
16.1 GB18.4 GBfits · 13.6 GB sparethe original weightsHugging Face →
Q8_0
GGUF · unsloth
8.7 GB11.0 GBfits · 21.0 GB sparepractically identical to the originalHugging Face →
Q6_K
GGUF · unsloth
6.8 GB9.1 GBfits · 22.9 GB sparevery close to the originalHugging Face →
Q5_K_M
GGUF · unsloth
5.9 GB8.2 GBfits · 23.8 GB spareclose; small differences in fine detailHugging Face →
Q4_K_M
GGUF · unsloth
5.0 GB7.3 GBfits · 24.7 GB sparegood; some loss in fine detail and textHugging Face →
Q3_K_M
GGUF · unsloth
3.9 GB6.2 GBfits · 25.8 GB sparenoticeable loss of detailHugging Face →
Q2_K
GGUF · unsloth
3.2 GB5.5 GBfits · 26.5 GB spareheavy loss; a last resortHugging Face →

“Needed” adds 1.5 GB of working memory for ERNIE-Image and a 0.8 GB reserve. Sizes 2026-10-01.