Captions for your images,
written on your Mac.
Loraset looks at your photos and writes the text you need: LoRA training captions, alt text, social posts, product descriptions and photo keywords. The AI runs on your Mac, so nothing is uploaded and there is nothing to pay per image.
Five modes, five jobs
Each mode has its own writing style, its own options and its own export formats. Projects stay separate, so a dataset never mixes with your shop photos.
Captions tuned per model
Comma-separated tags for SD 1.5, Pony and Illustrious, a phrase plus tags for SDXL, dense prose for FLUX, Qwen-Image, SD 3.5, HiDream, Wan and more. Trigger word, square crops and a Kohya-ready export.
Accessible descriptions
Short, medium or long image descriptions for websites, apps, e-mail and documents, in the style screen-reader users expect.
Ready-to-post captions
Instagram, LinkedIn, X, Facebook, TikTok and Pinterest, in the tone you pick, with optional hashtags and emoji.
Shop and marketing copy
Accurate descriptions for product, team and marketing photos. The AI never invents prices, sizes or materials — you give it the facts.
Titles, descriptions, keywords
For stock sites, Lightroom and photo libraries. Written straight into the photos' IPTC/XMP metadata on export.
Fix things in plain words
“Her eyes are green, not blue.” Loraset shows you what it would change, and updates every affected text only when you say so.
Built for real batches
A LoRA dataset is rarely three photos. Loraset is made for the hundred-image afternoon.
- Watch it write. Each text appears as it is generated, with progress and time remaining.
- Choose the detail level. 256 to 1536 px, with the real speed measured on your Mac.
- Pick your model. Qwen3.5 2B, 4B or 9B, or add any compatible MLX vision model from Hugging Face.
- Duplicate finder. Identical and near-identical photos, flagged; keep the largest copy in one click.
- Crop editor. Exact, non-destructive square crops for training exports.
- Bulk edits. Find & replace, prefix and suffix, a tag manager — all undoable.
- Every export. Images plus .txt files, Kohya folders, ZIP, CSV, JSON, or photos with embedded metadata.
- 12 languages. Write the texts in Spanish, French, German, Romanian, Japanese and more.
Private by design
Loraset runs an open-weight vision model locally with Apple's MLX framework. There is no server to send your photos to, because there is no server.
Your images stay put
Photos are copied into the app's own storage on your Mac and never uploaded. Your originals are never changed, moved or deleted.
One download, then offline
The only network request is the one-time model download. After that you can work with the Wi-Fi off.
Nothing collected
No accounts, no analytics, no tracking, no ads. Read the privacy policy — it is short.
Free, with a coffee jar
Every feature is unlocked for everyone: all modes, all models, all quality levels, the chat and every export format. Nothing is locked behind a purchase and there is no subscription.
If Loraset saves you an afternoon, you can buy me a coffee from inside the app: $5, $10 or $20, as often or as rarely as you like. It is a one-off thank-you and it unlocks nothing, because nothing is locked.
The AI runs on your Mac, so there is no cloud bill to cover. That is what makes free possible.
Requirements
- A Mac with Apple silicon (M1 or newer). The AI uses the Neural Engine and GPU through MLX.
- macOS 14 Sonoma or later.
- 8 GB of memory for the standard model; 16 GB or more for the largest one.
- About 3 GB of disk space for the model, plus room for your image copies.
Get Loraset for your Mac
Five ways to write for your images. Free, private, and available now.
Download on the Mac App Store Requires macOS 14 or later and a Mac with Apple silicon.