PixlStash
Local AI image manager for photo culling and dataset curation. Run it as a desktop app or self-hosted server, with on-device models and ComfyUI integration.
PixlStash is an open-source image manager for photographers, AI creators and people curating image datasets. It combines library search with tools for reviewing tags, ranking images and sending selected work through ComfyUI. You can use a desktop app or a self-hosted server with a browser interface.
Models run on your hardware, and library data stays in local storage. It uses no cloud API or external account. An internet connection is needed for the initial model downloads; subsequent starts use cached models. AI processing can use a CPU or NVIDIA GPU, with experimental support for AMD GPUs.
CLIP semantic search finds images by their content even when they aren't tagged. Face recognition groups people or characters, while reverse face and likeness searches find related images. Duplicate detection helps identify repeated files.
For tagging and captions, you can choose PixlStash Tagger, WD14, Florence-2 or JoyCaption. A review queue surfaces uncertain labels, and a tag health board helps assess automatic tagging. Quality, character-likeness and malformed-anatomy scores help with photo culling and generated-image review. Object segmentation is also available.
The ComfyUI connection works in both directions: PixlStash can run workflows on selected images, and ComfyUI nodes can search, load and save library images. Generated images return with tags and project, person and picture-set associations. A REST API supports other integrations.
Existing image folders can become libraries without moving, renaming or copying files. PixlStash supports separate libraries and folder exports. The software uses GPL-3.0; for face recognition, buffalo_l weights restrict use to non-commercial research, while AuraFace weights use Apache 2.0.
