Ernie Image 8B Low VRAM ComfyUI Workflow (6GB GPU) + One Click Installer

Ernie Image 8B is a newly released text-to-image diffusion model from Baidu designed to deliver high-quality, prompt-accurate image generation with strong visual coherence and detail. Compared to many existing image models, Ernie Image focuses heavily on improved text understanding, making it especially effective for complex prompts, multi-subject compositions, and stylistic control. It also introduces a more efficient architecture that performs well even when quantized, which is where this workflow really shines.
 
In this post, I’m sharing a very simple ComfyUI workflow built specifically for the quantized NVFP4 version of Ernie Image. This allows you to run the model on GPUs with as little as 6 GB of VRAM—no GGUF conversion required.
 
Alongside that, I’ve included a one-click installer that handles the setup process automatically, making it easy to get started with local AI image generation on low VRAM systems without dealing with complicated dependencies or manual configuration.
 
Ernie Image Base - https://huggingface.co/baidu/ERNIE-Image
Ernie Image turbo - https://huggingface.co/baidu/ERNIE-Image-Turbo
 
Preloaded Models Within the Installer (Low VRAM)
ernie-image-nvfp4.safetensors – Diffusion Model
https://huggingface.co/Bedovyy/ERNIE-Image-Quantized/tree/main
ministral-3-3b-fp8.safetensors – Clip Model
https://huggingface.co/fent67/ERNIE-Image-TEs-fp8_e4m3fn/tree/main
flux2-vae.safetensors – VAE Model
https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/vae/flux2-vae.safetensors
2xLexicaRRDBNet_Sharp.pth – Upscaler
https://huggingface.co/Thelocallab/2xLexicaRRDBNet_Sharp/blob/main/2xLexicaRRDBNet_Sharp.pth
The installer omits the standard Ernie Image 8B checkpoint models (BF16, FP8) to keep the download size smaller, but you can grab them anytime from the diffusion model link above.
 
Custom Nodes (Portable Windows Setup)
Custom Nodes w/ commands needed for Portable Windows package
ComfyUI Manager - https://github.com/ltdrdata/ComfyUI-Manager
Command to install requirements:
.\python_embeded\python.exe -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-Manager\requirements.txt
Use the ComfyUI Manager’s Install missing custom nodes feature to install any additional nodes that may not work initially after installation.
 
Speed & Performance
On mid-range systems, such as a 6 GB VRAM GPU with 16 GB system RAM, you can render a 1264x848 image in about 2 minutes using the nvfp4 quantized model. The workflow is optimized for low VRAM image generation, balancing GPU memory usage and compute load to maintain strong image quality, detailed textures, and consistent outputs even on limited hardware.
 
System Requirements
NVIDIA RTX 30XX / 40XX / 50XX GPU (FP16 supported)
CUDA-compatible GPU (minimum 6 GB VRAM)
Windows OS
At least 20 GB free storage
 
What’s Included
Portable ComfyUI Windows installer (pre-configured)
Automated model + custom node setup
Beginner-friendly layout with expandable advanced controls
 
Usage Notes
Enter a detailed text prompt describing your scene
For best results, expand your prompt using an LLM before pasting into ComfyUI
More descriptive prompts = better composition, lighting, and detail