Open-source Black Forest Labs alternatives

ComfyUI is an open-source application for creating images and other media with diffusion models. Instead of a fixed form, you build a workflow as a graph of connected nodes, where each node handles a step such as loading a model, encoding a prompt or sampling. It is written in Python on top of PyTorch and licensed under GPL-3.0.

The project provides a graphical interface, an API and a backend in one package. The modular design means workflows can be saved, shared and extended, and new capabilities are added through additional nodes. Because inference runs locally, users keep control over their models and outputs. The maintainers describe it more broadly as an engine for AI content creation.

The code is maintained by Comfy Org, which also runs the website at comfy.org, and releases are published on GitHub. ComfyUI suits artists, researchers and developers who want fine control over image generation pipelines and are comfortable wiring nodes together rather than using a simple prompt box.

InvokeAI, also called Invoke, is a creative engine for Stable Diffusion models aimed at professionals, artists and enthusiasts who want to generate and refine visual media. It runs a locally hosted web server with a React interface, and the project says it also serves as the base for several commercial products. The code is Apache-2.0 licensed, which the README describes as commercially friendly.

You download and install it on compatible hardware using the Launcher, then generate and iterate on images and build workflows. Features include a Unified Canvas for editing, and repository topics list text-to-image, image-to-image, inpainting and outpainting. It runs on Windows, macOS and Linux. Documentation covers installation, updates and tutorials, and the FAQ and Discord server handle common installation problems. Because generation runs locally, results depend on your GPU and the models you load.

SD.Next is an all-in-one web interface and server application for AI image and video generation. It is built on Stable Diffusion and supports dozens of models, and it is aimed at artists, researchers and hobbyists who want to create, refine, caption, upscale and process media from a single cross-platform application.

Its README lists features such as automatic model download from a list of reference models, automatic detection of models you add yourself and SDNQ, a quantization engine that can cut VRAM use by up to 4x. Balanced Offload shifts load between CPU and GPU memory so larger models fit on modest hardware. SD.Next can also caption and enhance images with more than 25 built-in language and vision models plus taggers, offers image processing and color-grading tools, and is localized into 15 languages.

SD.Next is written in Python with PyTorch and Hugging Face diffusers and is licensed under Apache-2.0. It performs platform-specific detection and tuning at install time, and it runs on your own computer or server.