SD.Next

Vladimir Mandic began SD.Next as a fork of the AUTOMATIC1111 web UI, and it has since grown into a separate application with its own installer, settings, and near-daily updates. It runs as a local server with desktop and mobile browser layouts and handles generation from prompts or source images, inpainting, outpainting, detailing, upscaling, captioning, and video. The supported list is long: Stable Diffusion 1.x through 3.5, SDXL, Stable Cascade, FLUX.1 and FLUX.2, Chroma, Qwen-Image, HiDream, Sana, Wan, Kandinsky, PixArt, and many more, and picking a reference model downloads it automatically. Its SDNQ quantizer and balanced offload between CPU and GPU memory let large models fit on cards with modest VRAM. You install by cloning the repository and running webui.sh, webui.bat, or webui.ps1; the launcher detects NVIDIA CUDA, AMD ROCm, or Intel XPU, and can be pointed at ZLUDA, DirectML, OpenVINO, ONNX, or Apple MPS instead. The code switched from AGPL to Apache 2.0 in July 2025, though every model keeps its own license terms. Suited to hobbyists and small studios who want one interface for many model families.