ERNIE Image Quants

Ernie in different quants:

• Mixed FP8 - Mostly fp8_e4m3 some are not quantized, fast.

• Mixed NVFP4 - NVFP4 except final layers to give a higher quality finish, faster than FP8

• NVFP4 - Mostly NVFP4 - Fastest

Note: You will only see speedups from NVFP4 on Blackwell series NVIDIA cards.

https://ernie.baidu.com/blog/posts/ernie-image/

text_encoders

• ministral-3-3b.safetensors

• ernie-image-prompt-enhancer.safetensors

vae

• flux2-vae.safetensors

Model Storage Location

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── ernie-image-turbo-nvfp4.safetensors
│   ├── 📂 text_encoders/
│   │   ├── ministral-3-3b.safetensors
│   │   └── ernie-image-prompt-enhancer.safetensors
│   └── 📂 vae/
│       └── flux2-vae.safetensors