The model is fp8 and based on Flux.1 D. It comes without VAE, Clip and textencoder to save disc space. So be sure to load those files in your workflow. You can find example workflows embedded in my posted images.This model can generate images that are NSFW - so be careful.After testing this model with different samplers and schedulers I really want you to use:Sampler: dpmpp_2mScheduler: betaSteps: 30Otherwise you are losing a lot of detail and that is were this model shines.Sure, using Euler simple with 20 steps is much faster and even using Hyper-Loras is possible - but then you can use any other model and get the usual mediocre stuff.So be patient and get some decent results with the settings listed above.For Clip I recommend using the new improved version from here:<a target="_blank" rel="ugc" href="https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/blob/main/ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors">https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/blob/main/ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors</a>

Files with SHA256: 39c40d2f7a514be6059fb58b680df48fd7ced2b4f14254d94e87cbe07d2805d8