Please see our <a target="_blank" rel="ugc" href="https://education.civitai.com/getting-started-with-stable-diffusion-3-5/">Quickstart Guide to Stable Diffusion 3.5</a> for all the latest info!<a target="_blank" rel="ugc" href="https://stability.ai/news/introducing-stable-diffusion-3-5">Stable Diffusion 3.5 Large Turbo</a> is a Multimodal Diffusion Transformer (MMDiT) text-to-image model with <a target="_blank" rel="ugc" href="https://stability.ai/research/adversarial-diffusion-distillation">Adversarial Diffusion Distillation (ADD)</a> that features improved performance in image quality, typography, complex prompt understanding, and resource-efficiency, with a focus on fewer inference steps.Please note: This model is released under the <a target="_blank" rel="ugc" href="https://stability.ai/community-license-agreement">Stability Community License</a>. Visit <a target="_blank" rel="ugc" href="https://stability.ai/license">Stability AI</a> to learn or <a target="_blank" rel="ugc" href="https://stability.ai/enterprise">contact us</a> for commercial licensing details.<h3 id="model-description-dtbl14w5u">Model Description</h3><ul><li>Developed by: Stability AI</li><li>Model type: MMDiT text-to-image generative model</li><li>Model Description: This model generates images based on text prompts. It is an ADD-distilled <a target="_blank" rel="ugc" href="https://arxiv.org/abs/2403.03206">Multimodal Diffusion Transformer</a> that use three fixed, pretrained text encoders, and with QK-normalization.</li></ul><h3 id="license-5t2rmj9h6">License</h3><ul><li>Community License: Free for research, non-commercial, and commercial use for organizations or individuals with less than $1M in total annual revenue. More details can be found in the <a target="_blank" rel="ugc" href="https://stability.ai/community-license-agreement">Community License Agreement</a>. Read more at <a target="_blank" rel="ugc" href="https://stability.ai/license">https://stability.ai/license</a>.</li><li>For individuals and organizations with annual revenue above $1M: Please <a target="_blank" rel="ugc" href="https://stability.ai/enterprise">contact us</a> to get an Enterprise License.</li></ul><h3 id="model-sources-qlaz41fpg">Model Sources</h3>For local or self-hosted use, we recommend <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI">ComfyUI</a> for node-based UI inference, or <a target="_blank" rel="ugc" href="https://github.com/huggingface/diffusers">diffusers</a> or <a target="_blank" rel="ugc" href="https://github.com/Stability-AI/sd3.5">GitHub</a> for programmatic use.<ul><li>ComfyUI: <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI">Github</a>, <a target="_blank" rel="ugc" href="https://comfyanonymous.github.io/ComfyUI_examples/sd3/">Example Workflow</a></li><li>Huggingface Space: <a target="_blank" rel="ugc" href="https://huggingface.co/spaces/stabilityai/stable-diffusion-3.5-large-turbo">Space</a></li><li>Diffusers: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-3.5-large-turbo#using-with-diffusers">See below</a>.</li><li>GitHub: <a target="_blank" rel="ugc" href="https://github.com/Stability-AI/sd3.5">GitHub</a>.</li><li>API Endpoints:<ul><li><a target="_blank" rel="ugc" href="https://platform.stability.ai/docs/api-reference#tag/Generate/paths/~1v2beta~1stable-image~1generate~1sd3/post">Stability AI API</a></li><li><a target="_blank" rel="ugc" href="https://deepinfra.com/stabilityai/sd3.5">Deepinfra</a></li></ul></li></ul><h3 id="implementation-details-82srlrhcl">Implementation Details</h3><ul><li>QK Normalization: Implements the QK normalization technique to improve training Stability.</li><li>Adversarial Diffusion Distillation (ADD) (see the <a target="_blank" rel="ugc" href="https://stability.ai/research/adversarial-diffusion-distillation">technical report</a>), which allows sampling with 4 steps at high image quality.</li><li>Text Encoders：<ul><li>CLIPs: <a target="_blank" rel="ugc" href="https://github.com/mlfoundations/open_clip">OpenCLIP-ViT/G</a>, <a target="_blank" rel="ugc" href="https://github.com/openai/CLIP/tree/main">CLIP-ViT/L</a>, context length 77 tokens</li><li>T5: <a target="_blank" rel="ugc" href="https://huggingface.co/google/t5-v1_1-xxl">T5-xxl</a>, context length 77/256 tokens at different stages of training</li></ul></li><li>Training Data and Strategy:This model was trained on a wide variety of data, including synthetic data and filtered publicly available data.</li></ul>For more technical details of the original MMDiT architecture, please refer to the <a target="_blank" rel="ugc" href="https://stability.ai/news/stable-diffusion-3-research-paper">Research paper</a>.

Files with SHA256: e8b1ddb7b11942eb9065cb009c0c591bf29d2037716d049ecd4e13465ec04637