SD.Next

SD.Next is a self-hosted web interface for artists, researchers and people who want to generate and edit images or videos on their own hardware. It builds on Automatic1111 WebUI's original codebase and supports Stable Diffusion alongside other diffusion models. It's open source under Apache 2.0.

Generation covers text-to-image, image-to-image and video creation from text or images. Editing tools include inpainting, outpainting and detail refinement, while LoRA, ControlNet and IPAdapters give users more control over the result. The same interface handles upscaling, frame interpolation, color correction and grading, so it covers work beyond the initial generation.

Captioning and tagging are built in, with LLM and vision-language models, OpenCLiP, WaifuDiffusion and DeepDanbooru. SD.Next can download models from its reference selection or detect models you supply. Custom scripts and extensions can add workflows.

Hardware flexibility is a main reason to consider it. SDNQ supports pre-quantized models and quantization during use to reduce GPU memory needs. Balanced Offload shares the memory load between CPU and GPU to fit larger models on limited hardware.

It runs on Windows and Linux with NVIDIA, AMD and Intel GPU support, and on macOS with Apple M1/M2 acceleration. CPU-only execution is also supported. Docker recipes cover CUDA, ROCm, Intel IPEX and OpenVINO. The browser interface has desktop and mobile layouts, multiple languages and selectable themes.