Open-source pixel-curator turns 2,000 raw images into a training-ready dataset locally
Impossible-Car-2809 · reddit · 2026-10-14
A Reddit user released pixel-curator, a local-first CLI that curates raw image/video folders into training-ready datasets for similarity training (identity, character, product, style), exporting directly to DreamBooth or Kohya LoRA formats.
Key claim: a curated 200-500 image subset outperforms a messy 2,000 for identity training, and curation is where most pipelines break down.
8-stage pipeline:
- Prefilter unusable files (resolution, corruption)
- Perceptual-hash dedup — often removes the majority of video-sourced sets
- CLIP embeddings + k-means clustering with map visualization to expose angle/lighting/composition imbalance
- Optical quality scoring incl. face-aware filtering
- CLIP text filtering to auto-strip watermarks/text
- WD14 tagging + Florence-2 captioning as .txt sidecars
- HTML report + spreadsheet round-trip for human keep/reject decisions
- One-command packaging into Kohya DreamBooth/LoRA directory structures
Fully local, no uploads, dry-run by default, resumable via SQLite manifest, runs on modest hardware.
Related event: pixel-curator: open-source CLI for LoRA dataset prep(4 posts)→
More from Multimodal
- 'Oh no!': AI video of a man chased by a giant monster goes viral — phusa29 · 2026-10-14
- ComfyUI-EdgeTTS v2.0.0 fixes Microsoft's 403 breakage, adds native SRT export — Narrow-Particular202 · 2026-10-14
- AI-generated portrait of Charlemagne impresses with detail — MeroveeFrancSalien · 2026-10-14
- Kroma: a free ComfyUI workflow chaining Chroma and Krea 2 with 2x upscale — gabrielxdesign · 2026-10-14
- Open-source Fish Audio Drama 3 studio lets you design character voices in any language — camenduru · 2026-10-14
- AI parody gives James Bond a tiny new partner: a milk-sipping puppy — FrameTwistCreates · 2026-10-14