PixlStash 1.0.0b2. A self‑hosted image manager for AI creators
A new self-hosted tool uses GPU inference to tag AI anomalies and organize thousands of generated images.
A new open-source tool aims to solve the organizational nightmare for AI image creators. PixlStash 1.0.0b2 is a self-hosted image manager built specifically for users of ComfyUI and other generation tools. It directly addresses the pain points of scattered folders, duplicate images, and lost workflows by providing a centralized hub. The application can monitor local folders, drag-and-drop images or ZIPs, and read critical metadata from ComfyUI outputs, allowing users to copy workflows back into the node-based editor with ease.
Beyond basic organization, PixlStash packs intelligent features powered by local GPU inference. It automatically tags images with descriptions and, crucially, identifies common AI generation flaws like 'Bad Anatomy' or 'Waxy Skin' using a fine-tuned ConvNeXt-base model. Creators can sort their libraries by visual similarity to a 'character,' group likenesses, and create curated sets for LoRA training. A standout feature is the ability to run ComfyUI's Image-to-Image and Text-to-Image workflows directly within the PixlStash GUI, with results automatically imported into the library. The tool is cross-platform, running on Windows, macOS, and Linux via PyPI, Docker, or a Windows installer, and emphasizes being a 'good AI citizen' with a configurable VRAM budget to leave resources for concurrent generation tasks.
- Automatically tags AI image anomalies like 'Flux Chin' using a local, fine-tuned ConvNeXt model with configurable VRAM usage.
- Runs ComfyUI workflows (I2I/T2I) directly inside its GUI and imports the results, bridging the gap between creation and management.
- Creates character-based likeness groups and exportable sets with captions, specifically streamlining the dataset preparation process for LoRA training.
Why It Matters
It professionalizes the AI image creation pipeline, turning chaotic folders into a searchable, actionable asset library for model training and refinement.