Features
Every feature, in one place
116 features, from the first photo to the final version. Each one says what it saves you – and whether it is in the first beta or comes later.
116 features
Restored your last filter.
{n} of {total} features
No feature matches. Try another word, or show all.
01
Organise photos
Bring photos in, see every one of them properly, and find the ones that need work.
28 ready in the first beta · 1 coming later
Drag photos in
Drop photos or folders on the grid, or paste with ⌘V. You see what will come in first, and exact copies are skipped.
Any folder layout
Point Skye Desk at the folders you already have, however they are organised.
Your originals never change
Skye Desk works on a safe copy. Every rename, delete and caption edit happens there, so your own folders stay exactly as they were.
Keep, set aside or reject
New photos wait in an inbox. Keep the good ones, reject the rest with a right-click – one ⌘Z brings any of them back.
Look at every photo properly
Fit or fill the grid, press Space to see a photo full size and use the arrow keys to move on.
Find any photo, in any project
Search by file name or caption in the current project or across all of them.
Rate and move on
Give a photo stars or a flag and the next one opens by itself, so sorting a hundred photos takes minutes.
Filters that work together
Combine stars, flags, colour tags and "has a caption" in one bar and see the count update as you go.
Smart folders
Save any view as a folder that fills itself – it stays up to date as you rate and tag.
Detailed filters
Filter by the number of tags or words in a caption, with AND, OR, NOT and wildcards.
Finder colour tags
Use the same colour tags as Finder – for example red for "don't train", green for "best of".
Health check
Before you train, Skye Desk marks photos that are too small, captions that are missing and photos that are almost the same.
Spot near-duplicates
Photos that are almost the same are marked, and you can open the pair side by side to choose one.
Quality score
Every photo gets a sharpness, brightness and contrast score, and each group of look-alikes shows the sharpest one to keep.
Angles and framing
See how many close-ups, half shots and full shots you have, and which angles are missing.
Dataset at a glance
Sizes, shapes, caption lengths and ratings as small charts – click a bar to see those photos.
Rename many at once
Give a batch of photos tidy names in one step, and undo it if you change your mind.
Swap a photo, keep its caption
Replace a photo with a better one; its caption stays put.
Recently deleted
Deleted photos go to Recently deleted first, so nothing is gone for good by accident.
Copy, pin and note projects
Start a new project from a copy of another, keep favourites on top and jot notes on each.
Start by saying what you make
A new project first asks what it is for – a product, clothing, a character, hair, a style, a video character or a motion – and sets up its folders, LoRA type and checks to match.
Back up a project
Pack a whole project – photos and captions – into one ZIP.
Crop preview
See which part of each photo the training will actually use, and move the crop by hand.
Cut out the subject
Models on your Mac – BiRefNet and SAM 2.1 – cut the subject, the clothes or the hair out of each photo as a mask. Nothing downloads until you agree.
Dataset versions
Save a version of your photos and captions, compare it with today and put old captions back.
Video clips in the dataset
Drop video clips into a dataset next to your photos. Skye Desk copies them in and reads their length, frame rate and size. Needs ffmpeg on your Mac.
Cut at every scene change
Long videos are cut into training clips where the shot changes, at the length and frame rate the trainer needs. The original is kept, and one ⌘Z undoes it. Needs ffmpeg.
Clip health check
Each clip is checked for length, frame rate, size and black frames, and the ones that need work are marked. Needs ffmpeg.
Photos and clips together
One dataset can hold both: photos teach the look, clips teach the movement, and a bar shows the mix.
02
Describe photos
Short descriptions teach the AI what changes from photo to photo. These tools keep them tidy.
10 ready in the first beta · 6 coming later
Captions that save themselves
Write next to the photo; it saves when you move on, and the trigger word is highlighted.
Change a word everywhere
Find and replace across every caption, preview the changes first, and undo the whole batch at once.
Trigger word tools
Add, remove or fix the trigger word (the made-up word that calls up your subject) in one go.
Rename a misspelt trigger word
Pick the old spelling, see every caption before and after, then apply – one ⌘Z undoes it.
Tag panel
Every tag with its count. Rename it, delete it or add it to the selected photos.
See which words take over
Tags sorted by how often they appear, with a hint when a word is on every photo or only once.
Put tags in order
Drag tags into the order you want; the trigger word always stays first.
Fill in many captions
Add the same words to the captions of every selected photo.
Length check
Each caption shows its word count and roughly how long it is for the model, with a hint when it is too long.
Captions match their photos
Photos without a caption and captions without a photo are found for you.
Tag suggestions
Suggestions from your project's own tags as you type.
AI writes the first draft
A caption model on your Mac – JoyCaption, Florence-2 or Qwen-VL – writes a first draft for every photo; you check and edit. Nothing downloads until you agree.
Tags written for you
A tagger on your Mac – WD14 or Florence-2 – suggests tags for every photo; you keep the ones you want.
Writing rules for each kind
Different rules for clothes, characters, hair and products, so the AI learns the right things.
Rewrite captions
Rewrite a batch of captions with a preview before anything changes.
Import captions written elsewhere
Bring in a set of captions from another tool, matched to photos by file name.
03
Set up training
Settings you understand, a check before you start, and the files your trainer needs.
9 ready in the first beta · 5 coming later
Presets
Five built-in starting points plus your own – apply, save or delete, and undo any of it.
Every setting tells you why
Each setting has one line explaining what it does, in plain words.
Form and config file stay in step
Change the form and the AI Toolkit config follows; edit the config and the form follows.
Check before you start
Photo count, captions, trigger word, words that shouldn't be there, duplicates and size are checked before a run.
Sample prompts written for you
Test prompts come from your captions, and one leaves out the trigger word to show whether the subject leaks everywhere.
Fit to your graphics card
Pick a GPU and see which settings change to fit its memory – apply them in one click.
Export for your trainer
Your photos and captions laid out the way each popular trainer expects, with a report of duplicates.
Training package
Everything for a run in one ZIP.
Time and cost before you start
An estimate of how long a run takes and what it costs on each GPU.
One-click cloud training
Rent a cloud GPU and start training without leaving the app.
Spending limit
Set a limit for each run; the GPU stops when training ends, sits idle or hits the limit.
Upload to cloud storage
Send your dataset to cloud storage with progress you can follow.
Live training progress
Watch the learning curve and new sample pictures while a run is going.
Licence check for base models
Each base model shows its licence at a glance: red when it forbids commercial use, amber when it comes with conditions, green when the licence allows commercial use – on Launch and in the Library. Always check the licence yourself.
04
Pick the final version
Compare what each training step learnt, without guessing, and keep the best one.
11 ready in the first beta · 3 coming later
Prompt × step grid
Every test prompt at every saved step in one grid. Zoom, hover to enlarge, hide steps you don't need.
Side by side, zoom together
Press C to put pictures next to each other; zoom and drag and they all move together.
Blind pick
Vote between two samples without knowing their step. After a few rounds you see which step wins.
Set the final version
Mark the winner as the final version in one click – and undo it if you change your mind.
Bring your own training back
Import the checkpoints and samples from AI Toolkit, OneTrainer or kohya into Runs, then pick the final version.
Checkpoint list
Each saved step with its size and samples – press F to make it final.
What changed between runs
A table of two runs' settings with the differences highlighted.
Runs side by side, step by step
Each run in a column at the same step; use the arrow keys to move through the steps.
Notes on each run
Write down what you tried and what you saw, right next to the run.
Export the grid
Save the step grid as a large picture or a PDF to share.
Export runs as a table
All runs in a spreadsheet file.
Download checkpoints
Bring finished checkpoints back from the training machine.
Strength and version grid
One prompt at several strengths and versions, side by side in one grid.
Publish your LoRA
Share what you made on Hugging Face or Civitai.
05
Generate
Keep prompts and results together, and send the good ones back into your dataset.
6 ready in the first beta · 4 coming later
Sort your results
Filter results by seed, prompt or note; rename, group and note them – all with undo.
Pass or reject
A to pass, X to reject, many at once, and ⌘Z to take it back.
Use these settings again
Put a picture's prompt back on the current shot, or copy its seed and settings.
Good results into the dataset
Add a result to your training photos in one click; duplicates are skipped.
Results arrive by themselves
Choose your ComfyUI output folder once; new pictures appear by themselves, sorted by shot.
A prompt for every shot
Keep one prompt and negative prompt for each shot you plan.
Generate without leaving the app
Send shots to ComfyUI and follow the queue.
Batch tests
Run many prompts and steps through one workflow at once.
Pictures ready for your shop
Sizes for each sales platform, white-background shots, one item in several colours, and a delivery pack to hand over.
Short dramas
Paste a script breakdown, keep a cast list and a shot board, and export numbered folders with a shot list (CSV).
06
Workflows and library
Good settings hide inside old pictures, and Skye Desk finds them. Your designs, references and resources sit next to the projects that use them.
12 ready in the first beta · 5 coming later
Read the workflow in a picture
Drop PNG, WebP or JPG pictures, .json files or a folder: the ComfyUI workflow inside is read and named.
Sorted by base model
Sorted by base model and structure, and filtered by what they do – upscaling, face fixes and more.
Export a workflow
Save a workflow as a file and open it in ComfyUI whenever you like.
Workflow templates
Keep favourite workflows as templates and copy them into projects.
Mark the key nodes
Mark which node loads your model and which one takes the prompt.
Ready-made recipes
Ready-made settings for generating and training with a given model.
My designs
Your own designs in themes – import, preview, rename and move them, with undo.
Reference boards
Boards for clothes, poses and more – drag pictures in and they are numbered for you.
Paste a screenshot
Press ⌘V to drop a screenshot straight onto a board.
Where it came from
Keep the source link and notes with each reference, and pin numbered notes on the picture.
Stars and days
Star favourites, group by day or month and download a whole group as a ZIP.
Link to a project
Attach references to the project they belong to.
Your resources list
A tidy list of the resources you use – add, rename, move, pin and note them.
Model files
Every model file with its type, family, licence and where it is used.
Videos in the library
Video thumbnails, hover previews and full-screen playback in your library.
Creator mode
A character profile and a series of posts for creators who keep one character across many pictures.
Download models
Download models from Hugging Face or Civitai without leaving the app.
07
Everywhere
The small things that make a work app feel calm: keys, undo, files and your language.
14 ready in the first beta · 2 coming later
Command palette
Press ⌘K and type: any page, project or command, right there.
Keys like Finder
Space to preview, Return to rename, ⌃1–⌃8 to switch pages, ? for the full list.
Find on the page
⌘F finds things on the page you are on.
Undo where it matters
In Dataset, Generate and Runs, every change can be undone with ⌘Z.
Right-click like Finder
Open, show in Finder, copy the file or its path – the same menu everywhere.
Panels your way
Show or hide the side panels and reset the layout; Skye Desk remembers.
Tasks and notes
Keep a to-do list on Home and on each project; overdue tasks turn red.
A gentle first start
The first time, Skye Desk asks your name and gets out of the way.
Step back in time
A list of recent changes – click one to go back to it.
Privacy mode
One shortcut blurs every picture – handy when you share your screen.
Send feedback
Help › Send feedback… (or ⌘K) shows the whole message first, then saves it as a file you send us. Nothing goes to a server.
Crash reports, off by default
Turn them on in Settings › Privacy. After a crash the app shows you the report first, with paths, file and project names, trigger words and keys removed. It stays on your Mac or is saved as a file – never sent by itself.
Ten languages
Skye Desk in your language – the same ten as this website.
Connections at a glance
One page lists ComfyUI, AI Toolkit, RunPod and the other services, and shows whether each one is set up. Live connections to RunPod and AI Toolkit come later.
Accounts and sync
An optional account that keeps your settings in step.
Several workspaces
Keep separate workspaces – one for each client, say – and switch between them from the menu.
What you can train
In the first beta, Skye Desk gets image and video datasets ready and checks them; you train with your own tool for now. Video clips need ffmpeg on your Mac.
Works with
- ComfyUIReads the workflow saved in your pictures and collects new results from its output folder.Ready in the first beta
- ComfyUISends your shots straight to it.Coming later
- AI Toolkit, OneTrainer, kohyaYour photos and captions laid out the way each trainer expects.Ready in the first beta
- RunPodRents a cloud GPU and trains for you.Coming later
Join the waitlist.
One email, the day the Mac download opens. Free during the public beta.