Dev vibecodes open-source suite: guided LoRA trainer, dataset captioner, image tournament tool
Speedyrulz · reddit · 2026-10-26
Reddit user Speedyrulz released a free, open-source Windows tool suite built entirely via vibecoding, now used daily for LoRA training and image management.
- RulzTrainer: guided-WebUI LoRA trainer for MOSS-Image, Bria FIBO 2.0 and Krea 2. A wizard recommends every setting based on your GPU memory and dataset; supports multi-GPU (data parallel or model split), int8/fp4 for 16GB cards, identity-similarity scoring for character LoRAs (body, not just face), live loss charts, and ComfyUI-compatible output.
- Rulz Training Captioner: captions datasets with local vision models via llama.cpp (any vision GGUF like Qwen-VL, Gemma 3, MiniCPM-V) or OpenAI-compatible servers; thumbnail grid with editable captions, multi-GPU captioning, trigger-word prefix/suffix.
- RulZ Image Tournament: pick best generations via side-by-side duels with synced zoom/pan; losers go to a Losers folder.
- RulZ Simple Video Clip Extractor: frame-accurate In/Out marking with lossless or NVENC export for video training data.
Author admits rough edges and welcomes contributions.
Related event: Developer vibe-codes free open-source local SD toolkit(2 posts)→
More from coding & agent
- 5-layer agent memory setup: Mem0 cuts tokens 93%, Snowflake ontology lifts accuracy 20% — Arindam_1729 · 2026-10-26
- Skills written entirely by the model are useless, and stacking skills taxes agent context — HamelHusain · 2026-10-26
- Grok Bot hands-on: auto-logs hundreds of tasks to Linear and runs scheduled security checks — koltregaskes · 2026-10-26
- agent-saifu: an MCP server that gives AI agents their own isolated DeFi wallets — DecisiveCS · 2026-10-26
- GigaMail ships three MCP security patches against indirect prompt injection leaks — Soft-Lie-434 · 2026-10-26
- Hermes Agent Lands Web Desktop UI, Enabled with a Single Command Locally — intellectronica · 2026-10-26