TagScribeR rebuilt: free local dataset studio with native LoRA training on AMD ROCm and NVIDIA
ArchAngelAries · reddit · 2026-10-06
Reddit user ArchAngelAries released a full modernization of TagScribeR, a free, open-source (GPL-3.0) local app that unifies image dataset building and LoRA training in one workspace.
Highlights
- One workspace: browse, filter, caption, edit and train on thousands of images
- Captioning with local Qwen3-VL, Gemma 4, JoyCaption, WD taggers, or any OpenAI-compatible server (LM Studio, Ollama); AI captions go through a human review queue with diffs
- Dataset tools: search-style filters, batch tag editing, duplicate/blur/resolution checks, privacy audit, training-ready export
- Native LoRA training: presets, adaptive learning rate, per-image loss watch, pause/resume, automatic memory planning (fp8/INT8/4-bit); supports Krea 2, Qwen Image 2.1, MiniMax H3, FLUX.2 Klein, SDXL family and more
- Cross-platform: native AMD ROCm on Windows/Linux, CUDA, CPU fallback
The training engine is a port of Peter Neill's Fizgig. Krea 2 training was verified end-to-end on an RX 7900 XT; other paths are covered by automated tests only. The upgrade was built with Claude Opus 5.5.
More from coding & agent
- dhh's AI-coded Campfire benchmark questioned: Rust got 398 tuning commits, Elixir got one pass — kristoph · 2026-10-06
- Ambient intelligence demo: Opus 5.5 agent shares your screen and knows what you're looking at — RileyRalmuto · 2026-10-06
- n8n Grew 10x Last Year Despite Being Declared Dead Three Times, CEO Says — aakashgupta · 2026-10-06
- Dev builds My AI Usage Canvas to turn local Copilot usage data into token trends and insights — DanWahlin · 2026-10-06
- Ethan Mollick: my AI workflow spans local Codex, cloud Dot and Claude tasks — explaining it is chaos — emollick · 2026-10-06
- Developer builds a tool that turns math equations into playable music — measure_plan · 2026-10-06