Vibe coded and made a Knights of New Order like free open sourced tool for proof-checking deepbooru tags
Check every image for a single tag in sequence without losing your place.
Deepbooru TagWalker Beta, released by developer Elliezrah under an MIT license, introduces a radically simpler approach to cleaning up image tag datasets. Most tagging tools are image-centric—you open an image and edit its tags. TagWalker flips that: you select a single tag, and the program walks you through every image in your dataset one by one, asking only whether the tag is correctly applied. The process is consistent, sequential, and automatically advances to the next image after each yes/no answer, preventing context loss or accidental skips.
The developer describes it as the tool they always wished existed, operating similarly to CivitAI's 'Knights of New Order' mini-game. This is their first-ever vibe-coding project, built using Qwen 3.6 27B Q4 on an RTX 3090. Despite the initial challenges of coding without prior experience, the resulting tool is functional and focused. By centering the workflow on tags rather than images, TagWalker makes large-scale dataset auditing far more efficient—especially for users managing thousands of images with consistent tagging across a single concept.
- Tag-centric workflow: pick a tag, walk through every dataset image with yes/no prompts.
- Inspired by CivitAI's 'Knights of New Order' mini-game for lossless sequential tagging.
- First vibe-coding project using Qwen 3.6 27B Q4 on RTX 3090; released under MIT license on GitHub.
Why It Matters
Eliminates context switching for large dataset tag auditing—ideal for ML training data cleanup and automated curation.