Dev vibe-codes a free open-source SD toolkit: LoRA trainer, captioner, image picker
Speedyrulz · reddit · 2026-10-26
Reddit user Speedyrulz built his own suite of free, open-source, Windows-local tools after getting frustrated with existing AI image tools missing features or being overly complicated — now they're his daily drivers for LoRA training and image management.
The suite includes five projects:
- RulzTrainer: a guided-WebUI LoRA trainer for MOSS-Image, Bria FIBO 2.0 and Krea 2. A step-by-step wizard recommends every setting based on your GPU memory and dataset, with explanations. Supports multi-GPU (data parallel or model split — author实测 1.7-2x speedup with two cards), int8/fp4 quantization for 16GB cards, identity similarity scoring for character LoRAs, live loss charts, direct ComfyUI loading, and vast.ai deployment.
- Rulz Training Captioner: captions datasets with local vision models (any vision GGUF via llama.cpp — Qwen-VL, Gemma 3, MiniCPM-V — or LM Studio/Ollama/OpenAI-compatible servers), with an editable caption grid, trigger-word prefix/suffix, and skip-already-captioned support.
- RulZ Image Tournament: pick your best generation via side-by-side duels with synced zoom/pan; losers move to a Losers folder, with undo and resume.
- RulZ Simple Video Clip Extractor: frame-accurate clipping for video training data, with fixed-length segments, aspect-ratio crop, lossless/NVENC export, and skin-tone color matching.
- RulZ Image Organizer: local image search and organization.
The author admits everything is 100% vibe-coded with rough edges, and welcomes bug reports and contributions.
Related event: Developer vibe-codes free open-source local SD toolkit(2 posts)→
More from coding & agent
- Kubernetes works for vLLM inference, but agent sandboxes needed Modal's custom stack at 1M+ concurrency — NielsRogge · 2026-10-26
- badlogic on his 2026 agent workflow: "I tell them what to do, I yell when they do it badly" — steipete · 2026-10-26
- Indie dev's trip planner fails 3 AI review rounds before making offline prices read-only — SeanOliver · 2026-10-26
- $21 of DeepSeek API wrote Molten: a Rust+Metal engine serving 300+ tok/s on M5 Macs — GabGarrett · 2026-10-26
- EMNLP 2026 Findings: "Ask or Assume?" turns coding agents into proactive help-seekers — najoungkim · 2026-10-26
- Purchasing agent in production: >95% recall, >90% accuracy, author asks how to push F1 higher — Ancient_Ad_916 · 2026-10-26