Free macOS App Concept Studio Streamlines Video-Model Dataset Captioning and Adapter Training
Ok-Republic764 · reddit · 2026-10-28
To recreate the finale of Grease with his parents as the young leads for his mother's 60th birthday, the author needed to train several adapters with consistent continuity — and built Concept Studio, a free, open-source native macOS app that covers the whole pipeline from dataset collection to adapter generation.
Key features:
- Import images/videos from a local folder, Apple Photos or a Hugging Face dataset; automatic duplicate detection and flagging of data ill-suited to training or likely to confuse the model.
- One-click AI review of dataset composition with advice (add more of a certain kind of image, preprocess, or leave as is).
- Crop, enhance, upscale and color-balance; define the concept to train and map it onto data (e.g. tag a character in multi-person images), then generate captions for the whole dataset in one click.
- Caption style is configurable via switchable instruction files, with an optional critic round where a different model reviews the captions; manually finalized captions are never overwritten.
- Publish captions plus training data to a private HF dataset or save locally; rent a cloud GPU and train from the app with a built-in TensorBoard dashboard, 40 supported models, some with recommended defaults.
- Finished adapters publish to a private-by-default HF repo or save locally.
The app is free and open source, requires Apple Silicon and macOS 15+; project files, editing and local face matching stay on the Mac, training runs on RunPod GPUs.
More from Multimodal
- AI band Techhalla reimagines Slipknot's 'Wait and Bleed' as a 1971 soul performance — Mother-Pear-2392 · 2026-10-28
- Inside the Hailuo H3 Max retro-MV workflow: 15-second chunks, one reference image, multi-shot prompts — techhalla · 2026-10-28
- Open-source pipeline turns Seedance AI video into real PBR textures on your 3D mesh — OdinLovis · 2026-10-28
- AI-generated cyberpunk short 'VHS Sirens' shows off consistent characters and neon city detail — brythehuman · 2026-10-28
- First-time creator makes full animated music video with 40 clips via MiniMax H3 — Capable-Insurance-58 · 2026-10-28
- Midjourney sref combo 138606468+1306896884 yields textured flat illustration — sergeantsref · 2026-10-28