PixlStash 1.0.0 release candidate
The AI image management tool's release candidate introduces major workflow features based on user feedback.
The developer behind PixlStash has launched release candidate 2 (1.0.0rc2) for the AI image management tool's first full version. Based on Reddit feedback, the update introduces a major new project system, allowing users to organize characters, sets, pictures, and documents under specific projects—ideal for focused work like custom ConvNeXt fine-tuning. A new fast-tagging workflow, born from the developer's own daily use, lets users quickly review and adjust auto-generated tags. This includes seeing low-confidence rejected tags, dragging tags between accepted/rejected categories, and using tag auto-completion. The update also adds anomaly tagging, where users can define which tags (like "Flux Chin" or "Malformed Teeth") are flagged in red.
Beyond core features, PixlStash 1.0.0rc2 brings significant quality-of-life and integration improvements. Users can now search and filter by ComfyUI LoRAs, models, and prompt text. The API has been cleaned up and now supports bearer tokens for easier integration, alongside a provided example fetch script. Better VRAM handling, a compact mode, and numerous new keyboard shortcuts (like 'F' for find and 'T' for tagging) round out the update. The developer notes the main hurdle to the final 1.0 release is improving the ConvNeXt-based auto-tagger for certain anomalies like "missing limb," but the new integrated workflow should accelerate this training process.
- Adds a project system for organizing AI art assets, characters, and training data by project.
- Introduces a fast-tagging interface with drag-and-drop, auto-complete, and visual anomaly tagging (e.g., 'Waxy Skin').
- Enhances integration with a cleaned-up API using bearer tokens and adds search/filter for ComfyUI LoRAs and models.
Why It Matters
It streamlines the chaotic workflow of managing and tagging thousands of images for AI model training and fine-tuning.