Warning: Although these quants work perfectly with ComfyUI - I couldn't get them to work with Forge UI yet. Let me know if this changes. The original non-k quants can be found HERE which are verified working with Forge UI.
[Note: Unzip the download to get the GGUF. Civit doesn't support it natively, hence this workaround]
These are the K(_M) quants for HyperFlux 16-steps. The K quants are slightly more precise and performant than non-K quants. HyperFlux is a merge of Flux.D with the 16-step HyperSD LoRA from ByteDance - turned into GGUF. As a result, you get an ultra-memory efficient and fast DEV (CFG sensitive) model that generates fully denoised images with just 16 steps while consuming ~6.2 GB VRAM (for the Q4_K_M quant). Also, the quality is pretty close to the original DEV model with ~30 steps.
It can be used in ComfyUI with this custom node. But I couldn't get these to work with Forge UI. See https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050 for where to download the VAE, clip_l and t5xxl models.
Advantages
Quality similar to the original DEV model, while requiring only ~16 steps.
Better quality and expressivity than the 8-step HyperFlux in general.
For the same seed, the output image is pretty similar to the original DEV model, so you can use it to do quick searches and do a final generation with the dev model.
Sometimes you might even get better results than the DEV model due to serendipity.
Disadvantage: requires 16 steps.
Which model should I download?
[Current situation: Using the updated Comfy UI (GGUF node) I can run Q6_K on my 11GB 1080ti.]
Download the one that fits in your VRAM. The additional inference cost is quite small if the model fits in the GPU. Size order is Q2 < Q3 < Q4 < Q5 < Q6. I wouldn't recommend Q2 and Q3 unless you absolutely cannot fit the model in memory.
All the license terms associated with Flux.1 Dev apply.
PS: Credit goes to ByteDance for the HyperSD Flux 8-steps LoRA which can be found at https://huggingface.co/ByteDance/Hyper-SD/tree/main
Description
FAQ
Comments (4)
And Q8?
The K(_M) quants appear not to work on Apple Silicon (M1, M2, etc). The HyperFlux 16-steps works fine.
I have been trying and comparing this flux Model and others for a long time, but this flux model is easy to use, has no problems when using many LoRa, and can produce good images. But I hope this flux model can be improved or updated again to be even better.