Instructions to use Comfy-Org/Ideogram-4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use Comfy-Org/Ideogram-4 with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Nvfp4 vs nf4
I think the nvfp4 is actually an nf4 model. I had a few people ask me why i posted it because they thought it was only for blackwell gpus, the original model file from ideogram says nf4 π
I made this nvfp4 version from the fp8 version, it's not fully 4 bit but keeps some more sensitive layers in fp8.
The nf4 is made by ideogram team, and it's not ComfyUI compatible.
Using NVFP4 for both models degrades quality quite badly. But as Kijai said in another discussion, using NVFP4 only for unconditional (and FP8 for base model) works quite a bit better.
I made this nvfp4 version from the fp8 version, it's not fully 4 bit but keeps some more sensitive layers in fp8.
The nf4 is made by ideogram team, and it's not ComfyUI compatible.
Your the man Kijai!
I was wondering how a nvfp4 was able to run on my 3070 lol i took the chance and ran it with success! Thank you for your work!
I made this nvfp4 version from the fp8 version, it's not fully 4 bit but keeps some more sensitive layers in fp8.
The nf4 is made by ideogram team, and it's not ComfyUI compatible.
Your the man Kijai!
I was wondering how a nvfp4 was able to run on my 3070 lol i took the chance and ran it with success! Thank you for your work!
Upcast to FP16 since neither NF4 nor FP8 have native support in Ampere gpus.