vladmandic/sdnext

★ 7,087PythonAudience · designerComplexity · 3/5Setup · moderate

Mindmap

mindmap
  root((SD.Next))
    What it does
      Image generation
      Video generation
      Image captioning
    Hardware support
      NVIDIA CUDA
      AMD ROCm
      Intel Arc
      Apple Silicon
    Key features
      SDNQ quantization
      Balanced RAM offload
      150 plus caption models
    Setup
      Run one script
      Auto dependency install
      Docker available

mindmap root((SD.Next)) What it does Image generation Video generation Image captioning Hardware support NVIDIA CUDA AMD ROCm Intel Arc Apple Silicon Key features SDNQ quantization Balanced RAM offload 150 plus caption models Setup Run one script Auto dependency install Docker available

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Generate AI artwork from text prompts using Stable Diffusion through a browser interface without writing any code.

USE CASE 2

Run large image-generation models on a mid-range GPU by using the built-in SDNQ quantization to cut VRAM usage by up to 4x.

USE CASE 3

Caption or tag a batch of existing images automatically using one of the 150+ bundled captioning models.

USE CASE 4

Create AI-generated videos through the same visual interface used for image generation.

Tech stack

PythonCUDAROCmDockerDirectML

Getting it running

Difficulty · moderate Time to first run · 30min

Requires a compatible GPU with appropriate drivers (CUDA for NVIDIA, ROCm for AMD), the install script handles Python dependencies automatically.

In plain English

SD.Next is an open-source web application that gives you a visual interface for generating images and videos using AI models, particularly those in the Stable Diffusion family. You run it locally on your own machine, point a browser at it, and use the visual controls to create AI-generated art, caption existing images, and apply processing effects without writing any code. The application supports a broad range of AI models for image generation and ships with a built-in installer that handles dependencies and updates automatically. It includes a desktop-optimized interface and a mobile-friendly version for use on smaller screens. A few features set it apart from similar tools. Its SDNQ quantization engine can reduce how much GPU memory a model needs by up to four times, making it possible to run larger models on hardware that would otherwise struggle. A balanced offload system dynamically moves data between CPU and system RAM and GPU memory, further stretching limited hardware. For describing or tagging existing images, the tool bundles over 150 captioning models and support for more than 25 visual language models. Hardware support is unusually broad. The application runs on NVIDIA GPUs via CUDA, AMD GPUs via ROCm or ZLUDA, Intel Arc GPUs, Apple M1 and M2 chips, and any device compatible with OpenVINO or DirectML. Docker images are available for the main GPU backends. Getting started requires cloning the repository and running one script (webui.sh on Linux and macOS, webui.bat on Windows). The script handles the rest of the installation. Documentation, changelogs, and a Discord community are linked from the project page for support.

Copy-paste prompts

Prompt 1

I have cloned SD.Next and run webui.sh. How do I load a custom Stable Diffusion model and generate my first image?

Prompt 2

My GPU only has 6GB VRAM. How do I use SD.Next's SDNQ quantization and balanced offload to run larger models without running out of memory?

Prompt 3

I want to use SD.Next on my AMD GPU with ROCm. What additional steps do I need after the base install?

Prompt 4

How do I access SD.Next's mobile interface from my phone while the server runs on my desktop?

Prompt 5

I want to batch-caption a folder of images using SD.Next. Walk me through selecting one of the captioning models and running it on a folder.

Open on GitHub → Explain another repo

← vladmandic on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.