Looks like no one quantized NetaYume Lumina v3, so don't mind if I do.

This model page contains fp8 scaled DiT. And a fp8 scaled Gemma 2 2b (the text encoder) too, just for convenience.

I chose the scaled fp8 because

• Comfyui supports it out-of-the-box, iirc, since May 2025.

• It can provide nearly identical quality comparing to the original model.

• 2.5Gb is small enough. No need to use gguf q6, q4 etc.

Tested on latest Comfyui v0.3.62.

All credit belongs to the original model author. License is the same as the original model.