🚀New Image Model🖼️ - Ernie Image Installer Update (Windows only) (Patreon)
Downloads
Content
Hi Friends,
I don't usually update the installer software twice in such a short time, but when a promising new image model comes along, I can't help myself — I have to put it to the test. This time, it's Ernie Image, and it's nothing short of impressive.
It's a ComfyUI Workload.
How does it compare to ZImage Turbo?
Quick takeaways:
Speed and photorealism — both models are equally matched
Style and text rendering — Ernie pulls ahead, making it a strong choice for infographics and artistic styles. It does produce occasional spelling errors, but still outperforms ZImage Turbo in this area
Human anatomy — Ernie's weak point, and particularly noticeable in 16:9 images
What's included today:
Updated installer software
Image and prompt examples (embedded in the Images drag & drop to ComfyUI)
The workflow for the full model
The workflows come with a built-in prompt enhancer. Honest verdict? I'm not a big fan. It can improve results, but in my experience it tends to be a one-trick pony rather than a reliable tool.
BUT! THERE IS A BUT! For Text In Images it seems to have less typos when activated. So I would not delete it from my workflow.
Recommended settings:
Best resolutions:
1024×1024
848×1264 / 1264×848
768×1376 / 1376×768
896×1200 / 1200×896
Parameters:
Guidance scale: 1.0
Inference steps: 8
Want the full model?
You can download the full Ernie Image model here:
🔗 https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/diffusion_models/ernie-image.safetensors
When using the full model, set CFG to 4.0 and steps to 40–50.
also use my "This is not the Turbo Workflow this is Ernie NORMAL-24GB Needed.json" Workflow.
Coming in May — The Diffusion Model Almanac(A PDF Guide)
I've been working on something I'm genuinely excited about: a comprehensive guide designed to cut through the confusion that has built up around modern diffusion models.
P.S.: I know it's supposed to be "almanac" without the 'h' I saw that too late 🤗When Stable Diffusion 1.5 launched, things were relatively simple — Flux, Pony, and the rest didn't exist yet. Now the landscape is far more complex, and many of you have told me the basics feel overwhelming. This guide is my answer to that.
The first chapters will be free for everyone, for two reasons:
I want this knowledge to be widely accessible
It will serve as the foundation for all future guides
Now go create something amazing. 🚀
Chris