Hyper-SD is one of the new State-of-the-Art diffusion model acceleration techniques. In this repository, we release the models distilled from SDXL Base 1.0 and Stable-Diffusion v1-5Project Page: <a target="_blank" rel="ugc" href="https://hyper-sd.github.io/">https://hyper-sd.github.io/</a>Original Repo: <a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/Hyper-SD">https://huggingface.co/ByteDance/Hyper-SD</a>Original Paper: <a target="_blank" rel="ugc" href="https://arxiv.org/abs/2404.13686">https://arxiv.org/abs/2404.13686</a>

ByteDance Hyper-SD XL

Hyper-SDXL-8steps-lora.safetensors

Hyper-SDXL - 8 Steps

Hyper-SDXL - 8 Steps CFG

Hyper-SDXL - 4 Steps

Hyper-SDXL - 2 Steps

Hyper-SDXL - 1 Step

Lightning: <a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/SDXL-Lightning">https://huggingface.co/ByteDance/SDXL-Lightning</a> Hyper: <a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/Hyper-SD">https://huggingface.co/ByteDance/Hyper-SD</a> These are LoRAs designed to generate images with fewer steps. These LoRAs are owned by ByteDance. I have uploaded them, so we can use them for on-site generation. 1) Use a CFG between 1 to 2. 2) Use at least 8 steps. 3) Higher CFG requires more steps. 4) Some models only seem to work with Eular A when using this LoRA. 5) Use SGM Uniform or Trailing for the Scheduler. 6) If you notice weird yellow/green blotches on the image it means you need to either raise your steps or lower your CFG. 7) If you download this LoRA I recommend merging/mixing your checkpoint with this LoRA as it improves performance and can give back control of CFG. 8) CFG doesn't work with this LoRA on the Drawthings app unless you merge/mix your checkpoint with it.

Lightning/Hyper - 8step

<a target="_blank" rel="ugc" href="https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-SDXL-8steps-lora.safetensors">https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-SDXL-8steps-lora.safetensors</a>

Files with SHA256: ca689190e8c46038550384b5675488526cfe5a40d35f82b27acb75c100f417c1