Self-Hosted LoRA Dataset Workbench
Ill-Ant-9489 · reddit · 2026-07-13
The author open-sourced a self-hosted tool, LoRA Dataset Studio, aiming to integrate the entire workflow—from a single reference image to a usable LoRA—into one interface.
The emphasis is on pre-training dataset construction rather than the training itself, featuring:
- Generating reference image variants and filtering them using composite metrics
- Auto-captioning and categorizing compositions (face / half-body / full-body)
- Filtering out off-identity images using face similarity scores
- Supporting crawling for concept/style datasets, with deduplication and quality filtering
- Integrating ai-toolkit for training, providing checkpoint evaluation and ranking
The tool supports three dataset types: character, concept, and style. It can run API-only or locally with ComfyUI and ai-toolkit. The project is MIT-licensed, fully self-hosted, with no accounts and no telemetry.
Related event: LoRA Dataset Studio Unifies Photo-to-LoRA Workflow(2 posts)→
More from Multimodal
- Ultimate Face Fix: Open-Source Multi-Face Repair Node for ComfyUI — Merserk13 · 2026-07-22
- Getting Started with AI Video: Solving Consistency and Censorship — cynicalnewenglander · 2026-07-22
- Storyboard-first workflows are making AI dance videos and influencers more consistent — aftahi_ai · 2026-07-22
- Runpod MCP and Claude help spin up image and video generation workflows — 802high · 2026-07-22
- Midjourney prompt turns a bee into a glitching pixel explosion — michaelrabone · 2026-07-22
- A physics reward can improve video generation without creating a real physics engine — Dapper-Drawer4546 · 2026-07-22