You can deploy this model and start earning money today!

realrebelai/
NL-Diffusion-Image_GGUF

Name: NL-Diffusion-Image_GGUF
Availability: InStock

nvidia/NL-Diffusion-Image

quantized

Views

No views yet

Rebels NL-Diffusion-Image GGUFs

ComfyUI_Rebels_NLD Custom Nodes

GGUF loader + text-to-image nodes for NVIDIA NL-Diffusion-Image (masked discrete diffusion LM + IBQ VQ decoder) on consumer hardware. By RealRebelAI.

https://github.com/RealRebelAI/ComfyUI_Rebels_NLD

NODES ARE OPERATIONAL BUT SLOW. Currently working on patches for speed ups. they will run in their current state but i recommend git pulling frequently.

Install

Clone into ComfyUI/custom_nodes/.
Requires the city96 ComfyUI-GGUF fork in the same custom_nodes/ folder (used for dequant).
Put the model files (dropdown-selected, no paths):
- dLM GGUF → ComfyUI/models/unet/
- vqvae (bf16 .safetensors) → ComfyUI/models/vae/
The config/tokenizer/modeling code ships in model_assets/

IMPORTANT!

model.safetensors file MUST go in "custom_nodes\ComfyUI_Rebels_NLD\model_assets\emu3_vqvae"

https://huggingface.co/nvidia/NL-Diffusion-Image/blob/main/emu3_vqvae/model.safetensors

Nodes

NL-Diffusion dLM Loader (GGUF) — pick gguf_name and vqvae_name from dropdowns, choose device.
NL-Diffusion Text to Image — prompt, size, steps, guidance, temperature, seed → IMAGE.

Notes

The dLM generates discrete token indices; the vqvae decoder turns them into pixels. It is not a latent VAE — it loads through this pack, not ComfyUI's VAELoader.
Vocab embeddings use row-gather dequant, so the 131k-row tensors never fully materialize.
Vision-tower (image-understanding / edit) weights are left on meta and not needed for t2i.

License

Model is under the NVIDIA One-Way Noncommercial License (research/development only). Quants inherit those terms — publish as license: other with the upstream terms linked.

realrebelai/NL-Diffusion-Image_GGUF