NitroFusion2 fast image generator in 1-2 steps based on Z-Image-Turbo I don't have high hopes for it. There's just a funny story behind it. Nitro Fusion 1 came out two years ago and was an unstable image generator. And Nitro Fusion 2 is like a finely tuned Z-Image Turbo. However, the name remains the same. And I have only one question: if Z-Image Turbo seems to be dead and out of date, then why make Nitro Fusion 2 or 3? For the two-stage mode, the authors released a special LoRA adapter for de-refinement (used with a weight of 0.5 in the first stage and disabled in the second stage to eliminate artifacts), as well as a series of quantitative GGUF weights with capacities from 2.0 GB (Q2_K_4XS) to 7.02 GB for a ready-made workflow (Q8-UI) and made convenient. Tell me, is this the kind of open news you need? Should I leave it on the channel or not? - Leave - No need Sources: GitHub project page Hugging face arXiv.

NitroFusion2 fast image generator in 1-2 steps based on Z-Image-Turbo I don't have high hopes for it. There's just a funny story behind it. Nitro Fusion 1 came out two years ago and was an unstable image generator. And Nitro Fusion 2 is like a finely tuned Z-Image Turbo. However, the name remains the same. And I have only one question: if Z-Image Turbo seems to be dead and out of date, then why make Nitro Fusion 2 or 3? For the two-stage mode, the authors released a special LoRA adapter for de-refinement (used with a weight of 0.5 in the first stage and disabled in the second stage to eliminate artifacts), as well as a series of quantitative GGUF weights with capacities from 2.0 GB (Q2_K_4XS) to 7.02 GB for a ready-made workflow (Q8-UI) and made convenient. Tell me, is this the kind of open news you need? Should I leave it on the channel or not? - Leave - No need Sources: GitHub project page Hugging face arXiv.