The missing link in the breeding loop: turn your Redraw Live picks into a training-ready LoRA dataset —
right here, in your browser, with no sign-up and no GPU. When the dataset is done, it plugs straight into the free
Kaggle trainer to cook your LoRA.
Step 1 — name your trigger word
One made-up word the LoRA will learn as "your thing", e.g. jessfox, emberlibrarian. It gets stamped on every caption.
Step 2 — drop in your picks
Up to 30 images. They get center-cropped square and resized to 1024×1024 — the exact diet SDXL likes.
waiting for pictures
Step 3 — describe them (your words work best)
Type a caption under each picture — a plain description like "a red fox librarian shelving glowing books, painterly".
The ✨ button can draft one for you with an on-device model (first run downloads it — Wi-Fi recommended, and it shows its progress).
add pictures first
Step 4 — forge the dataset
Downloads one zip: numbered images + matching .txt captions + a dataset card. Upload that zip as a Kaggle dataset
and point the free trainer at it.
The whole pipeline, end to end
Redraw Live — breed ten versions per edit, tap favorites onto your Picks shelf.
This Forge — caption, crop, and zip the picks into training data.
Kaggle trainer (free GPUs, ~30 hrs/week) — trains your LoRA from the zip. The notebook's already built and waiting.
Your LoRA — jessfox becomes a style/character you can summon anywhere SDXL runs.