Skye DeskNext beta
Coming in the next beta
Built and being tested. Not in the current build yet.
Steadier and faster
- Found and fixed four causes of rare crashes. In our tests, repeated full checks of the test build then ran without a crash.
- In our tests with 50,000 images, opening a 500-image dataset paused the window for 0.1 seconds instead of 1.4, and searching from Home took 0.02 seconds instead of 0.45.
- Fixed a problem where a project’s trigger word could be saved again when you only clicked away from the box. It is now saved only when you type in it.
- Switching pages and projects is smoother: the old page fades out while the new one slides in, and a project switch no longer freezes the window – in our tests the longest pause dropped from about 1 second to 40 milliseconds. With “Reduce motion” on in macOS, moves become short fades.
New features
- A dataset that works like FinderPress Return to rename, ⌘⌫ to move to the Trash (with undo), and use the same colour tags as Finder.
- Drop or paste anywhereDrop onto the grid, a project in the sidebar, the Projects page or the Home page, or paste with ⌘V.
- Batch renameChoose a pattern and see every new name before you apply it.
- Dataset health checkImages missing a trigger word, too small for training, oddly shaped or almost identical to another are marked, with a fix next to each.
- Search every project from HomeSearch all projects by caption, file name or prompt.
- Tag panel for the whole datasetSee every tag and how often it is used, then add, remove or rename it on many images at once, with undo.
- Check AI captions one by oneSee each old and new caption side by side and keep only the ones you tick.
- Trigger word spelling checkDifferent spellings of your trigger word (the word that calls up your subject) are found and fixed in one step.
- Training presets and checklistPick the kind of LoRA for ready-made settings and a checklist before each run.
- Workflows found in your imagesSkye Desk reads the ComfyUI workflow saved in your images and groups them by base model (the AI model your LoRA adds to).
- Use a prompt as a caption draftInsert the prompt saved in an image into its caption, then edit it.
- Compare runs and set the final LoRAOn the Projects page, compare two training runs side by side and mark one as the final LoRA, with its base model shown.