What is a LoRA?Get started
A LoRA is a small model that remembers what your thing looks like.
Ask an image AI for “my red coat” ten times and you get ten different coats. It has never seen yours.
A LoRA fixes that. You show it 15 to 40 photos of your coat, it learns the coat, and from then on the AI draws that coat – in any scene, any light, any pose.
Skye Desk is the Mac app that takes you from a folder of photos to that LoRA: it sorts the photos, helps you describe them, checks everything before training and helps you pick the best version.
A small add-on model that learns one subject – a coat, a character, a product, a style – from your photos. With it, an AI image model draws that subject the same way every time.
The big AI image model your LoRA adds to. A LoRA works with the model family it was trained on.
A made-up word, such as my_coat, that calls up your subject in a prompt.
The sentence that describes one photo. The AI learns from each photo and its caption together.
Your photosGet started
How many to take, and how to shoot them with your phone.
How many
Change everything except the subject
On your phone
Onto your Mac
More is not always better. Fifteen different photos teach more than fifty from the same spot.
Angles: front, three-quarter, side, back, and a few close-ups.
Backgrounds: three to five different ones. A background that is in every photo gets learnt too.
Light: window light, shade, a lamp – not all the same.
Distance: wide, medium and close.
Use the main camera at 1×, not the ultra-wide lens.
Turn off portrait mode, filters and beauty effects.
Wipe the lens and tap to focus on the subject.
Leave out other brands’ logos and any text you do not want learnt.
Train a real person only with their written consent, and never anyone under 18.
Your first LoRA, start to finishGet started
The whole route once, so you know where you are going.
Starting a cloud training run from inside Skye Desk is coming later. Until then you train with your own tool, such as AI Toolkit or OneTrainer.
Ready in the first beta Runs › Import training… brings in the checkpoints and sample pictures from AI Toolkit, OneTrainer or kohya, so you can compare them and pick the final version here.Train with your own toolDon’t want to use a cloud GPU?
Install, update and uninstallGet started
Put Skye Desk in Applications, open it the first time, keep it up to date or remove it.
Install
If macOS will not open it
Update
Uninstall Skye Desk
System requirements will be published with the first version.
The first time you open an app from the internet, macOS checks it. If it asks whether you are sure, click Open.
If it says it cannot check the app, open System Settings › Privacy & Security, scroll down, click Open Anyway and confirm.
You do this once for each new version.
Your projects, photos and settings are not inside the app, so they stay as they are.
Your workspace, your photos and your exports are ordinary folders. They stay where they are; delete them yourself only when you no longer need them.
Coming later
Updates from inside Skye Desk.
Create a projectOrganise photos
Tell Skye Desk what you want to make, and it sets the project up for that.
Video character and Video motion or camera move projects get a Clips section for video. The clips are kept in the clips folder of the project. Export for a trainer takes only the pictures and captions for now, so give the clips to your training tool straight from that folder.
Drama series is not a LoRA project: it opens Library › Creator, where you make character profiles, series, episodes and shots.
Train with your own tool
Add photosOrganise photos
Bring a folder of photos into a project.
Open a folder you already have
Video clips
Skye Desk reads PNG, JPG, WebP and BMP, and turns HEIC and TIFF photos into PNG as they come in. A caption is a .txt file with the same name as its photo.
Already keep your photos and captions in another layout? Open the folder as a project: Skye Desk recognises 20 common layouts, shows you a preview first, and a short wizard helps with any other layout.
Projects for a video LoRA have a Clips section on the Dataset page. It needs ffmpeg installed on your Mac: Skye Desk uses it for thumbnails, lengths and cutting.
Click Import clips… and choose a folder of videos (MP4, MOV, M4V or WebM). They are copied into the project; the originals stay where they are.
Cut into training clips cuts at every shot change and into pieces a trainer can use, with frame counts of 4n+1. ⌘Z puts the originals back.
Every clip is checked: too short, longer than one training clip, a different frame rate, a short side under 480 px, black picture, or a file that cannot be read.
In a video character project, a bar shows how many pictures and clips you have against the aim for that kind of project.
The clips are kept in the clips folder of the project. Export for a trainer takes only the pictures and captions for now, so give the clips to your training tool straight from that folder.
Coming later
Turning an existing folder into a project in place.
Look at a photo up closeOrganise photos
See one photo full size without leaving the app.
Rename, tag and tidyOrganise photos
Work with photos the way you do in Finder.
Your originals stay untouchedOrganise photos
Skye Desk works on a safe copy, so trying things costs nothing.
Dataset versions
The top bar says “Safe copy · originals untouched”. Every rename, delete and caption edit happens in the copy; your original photos are never changed.
Deleted photos wait in Recently deleted until you empty it.
Changes to many photos show a preview first, and one ⌘Z undoes them all.
Photos and captions are ordinary files. Back up your work folder like any other folder.
Coming later
Rolling the whole set of photos back to an old version.
Seeing in Runs which version a training run used.
Where your files liveOrganise photos
Your work is ordinary folders on your Mac: what is where, how much space it takes and how to back it up.
The safe copy
How much space
Back up
Move to a new Mac
Skye Desk works in one folder, the workspace. Each project is a folder in it: your photos, with a .txt caption of the same name next to each one. Skye Desk keeps its own settings and notes in a folder called _desk inside the workspace.
Settings › General shows the workspace folder, and the folder every export starts in.
Skye Desk works on a safe copy, and the top bar says Safe copy. Every rename, delete and caption edit happens in the copy; your original folder stays as it is.
Your photos take the same space as before.
Every export is a full copy of the photos in it, saved in Downloads unless you choose another folder in Settings › General.
AI models on your Mac, when they arrive, need about 3.6 GB for the first set.
On Projects, right-click a project and choose Back up as ZIP…: the whole project folder, photos and captions, in one file, saved in your export folder.
Coming later
Choosing which folder Skye Desk opens, for example one you have just moved.
Check your photos before trainingOrganise photos
The Health check finds the photos that would make your LoRA worse.
Quality scoreOrganise photos
Find the blurry, dark and washed-out photos in one go.
Stats on the same toolbar shows sizes, shapes and caption lengths; click a bar to see those photos.
Smart filters and foldersOrganise photos
Get to the photos that need work, and come back to them later.
Angles and framingOrganise photos
See whether your photos cover enough angles.
Shots reads your captions. Photos whose captions do not say the angle are left out.
How captions workWrite captions
One sentence per photo, starting with your trigger word.
Describe what should change from picture to picture. Leave out what the LoRA should learn.
Change a word in every captionWrite captions
Fix one phrase across all your captions at once.
Trigger word toolsWrite captions
Keep your trigger word spelt the same everywhere.
More on the Trigger tab
With Find & replace
Add where missing puts the trigger word at the start of every caption that has none. It does not fix a misspelt one.
Move to the start moves the word to the front of each caption; Remove takes it out.
Find & replace works for any text. On its tab, find the misspelt word (vel_trench), replace it with your trigger word (vell_trench) and tick Whole words, then check the preview and click Apply.
Try it
vell, trench coat, studio light, plain background
Vell, trench coat on a rainy street at night
vel, trench coat, close-up of the collar
1 / 3 captions use it exactly · 2 spelt another way
Tags and balanceWrite captions
See which words take over your captions.
If one background or pose is in most captions, the LoRA will learn it too. Rewrite those captions or shoot more variety.
Write many captions at onceWrite captions
Add the same words to a group of photos.
Coming later
AI caption drafts written on your Mac.
Importing captions from another AI.
Reviewing new captions one by one.
AI models on your MacWrite captions
Small AI models that draft captions and tags and cut masks on your Mac, without uploading your photos.
Which models
Download and remove
What needs them
Coming later
None of this is in the first beta. It is here so that you can plan the disk space and check the licences.
Skye Desk offers a small first set: about 3.6 GB in all, with the Python they run in (about 1.5 GB).
Larger models are listed too, for when you have the space:
JoyCaption follows the Llama 3.1 Community License, which adds conditions – credit, naming and an acceptable-use policy. Read the licence text before you use it. Models whose licence does not allow commercial use, such as RMBG-2.0, Qwen2.5-VL-3B and the clothes and face parsing models, are not offered.
Nothing downloads until you confirm. First Skye Desk shows each model’s name, size and licence, and how much free space will be left.
Models come from Hugging Face. A download can be stopped and resumed, and every file is checked before it is used.
They live in their own folder, ~/Skye Desk/models, outside your workspace and with their own Python: nothing else on your Mac changes.
To free the space, quit Skye Desk and delete the model’s folder there.
Auto caption on the Dataset toolbar writes drafts for the photos you choose. They are a preview: nothing reaches your captions until you click Apply, and ⌘Z undoes the whole batch.
Masks cuts out the clothes, the face or the hair, so that training can focus on them. Masks of the whole subject need BiRefNet from the larger set.
Without the models everything else works as before, and you write the captions yourself.
Captions for each kindWrite captions
What to write for clothing, characters, hair, products and styles.
Clothing
Characters
Hair
Products
Styles
my_coat, a trench coat on a dress form, side view, plain grey background
Describe: the pose, the person or dress form, the background, the light, the camera angle.
Leave out: the fabric, colour, cut and details you want the coat to keep – the LoRA should learn those.
my_char, sitting on a bench, three-quarter view, soft evening light, city street
Describe: what the character is doing, the pose, the clothes if they should change, the place and the light.
Leave out: the face, the hair, the colours and markings that make the character who they are.
my_hair, woman seen from the side, indoor window light, plain wall
Describe: who wears it, the angle, the light, the background, the clothes.
Leave out: the cut, the length, the colour and the texture of the hair itself.
my_bottle, on a stone plinth, top view, morning light, soft shadow
Describe: where it stands, the angle, the light, the props and the background.
Leave out: the shape, the label, the colours and the logo – the parts that must stay the same.
my_style, a lighthouse on a cliff, wide view
Describe: the subject of each picture – it should be different in every one.
Leave out: how it looks: the brush, the palette, the line – that is what the LoRA learns.
Set up a training runTrain and choose
Pick a starting point and see why each setting is what it is.
The check before trainingTrain and choose
Catch the problems that would waste a run.
Export for your training toolTrain and choose
Get the dataset into the layout your trainer expects.
Train with your own toolTrain and choose
The first beta does not train by itself: export the dataset, train in AI Toolkit, OneTrainer or kohya_ss, and keep the best result.
AI Toolkit
OneTrainer
kohya_ss
After training
Skye Desk prepares and checks the dataset; the training runs in a tool you already use, on a computer that can train. Each export is a new folder with your photos, their captions, duplicates.csv and HOW TO USE.txt, which names the exact folder to point your tool at.
What the export looks like
Launch writes an AI Toolkit config from the settings you chose. Click Copy next to config.yaml, paste it into AI Toolkit and change folder_path to your export.
Your tool saves a .safetensors file every few hundred steps, and sample pictures if you asked for them. Look at the samples, keep the step that looks most like your subject and use that file in ComfyUI.
Use your LoRA in ComfyUI
Coming later
Bringing your tool’s checkpoints and sample pictures into Runs, to compare them and set the final version there.
Training in the cloudTrain and choose
Rent a GPU and train without leaving Skye Desk.
This is the part that is not ready yet: starting the run, watching it and stopping the GPU all happen in Skye Desk later. You will need an account with a GPU service such as RunPod.
Coming later
Start a cloud run from Launch.
A spending limit, and the GPU stopping by itself when the run ends or sits idle.
Live progress and the loss curve.
Downloading checkpoints from the GPU.
Don’t want to use a cloud GPU?Train and choose
Ways to train without renting a GPU, and to pick the result in Skye Desk.
Your own tool, today
Bring the results back
Your own NVIDIA computer
On this Mac
Export the dataset for AI Toolkit, OneTrainer or kohya_ss, train wherever that tool runs and use the result in ComfyUI.
Train with your own tool
In Runs, choose Import training… and pick the output folder of AI Toolkit, OneTrainer or kohya. The checkpoints and samples come in as a run, so you can compare them and pick the final version.
Coming later
Starting a run from Skye Desk on your own NVIDIA computer, through AI Toolkit.
Training on the Mac itself is being evaluated. Nothing is decided, and there is no date.
Pick the final versionTrain and choose
Compare the steps of a run and keep the one that looks most like your subject.
Ready in the first beta Runs › Import training… brings in the checkpoints and sample pictures from AI Toolkit, OneTrainer or kohya, so you can compare them and pick the final version here.Train with your own toolDon’t want to use a cloud GPU?
Compare runsTrain and choose
See what changed between two training runs.
Coming later
Bringing the checkpoints and sample pictures from your own training tool into Runs.
Live progress of a running training.
Strength-by-version grids made in ComfyUI.
Use your LoRA in ComfyUIUse your LoRA
Where the file goes, which node loads it, how strong to set it and where the trigger word goes.
Put the file in place
Load it
Strength
The trigger word
Your training tool saves the LoRA as a .safetensors file, usually one for every few hundred steps. Use the step you chose.
Load LoRA has two strengths, strength_model and strength_clip. Give both the same value to begin with.
Your subject is weak or missing: raise the strength a little, up to about 1.2.
Pictures look stiff, burnt or all alike: lower it to 0.5–0.7.
Change it in steps of 0.1 and keep the seed the same, so that you only see what the strength does.
my_coat, a woman walking on a beach at sunset, wide view, soft evening light
Base model licences differ. Some do not allow selling what you make; check before you sell or publish.
Can I use it commercially?
Connect ComfyUIUse your LoRA
Set ComfyUI’s output folder once, and the pictures you make come into Skye Desk by themselves.
Set it up
File pictures under a shot
Good to know
Skye Desk watches the folder ComfyUI saves its pictures in, and brings each new one into the project you have open on Generate.
Start the file name with the shot’s name: in ComfyUI’s Save Image node, set filename_prefix to it. Point at a shot on Generate to see its name. Pictures whose names match no shot still come in, without a shot.
Only pictures made after you set the folder come in. Older ones stay where they are.
Nothing in the ComfyUI folder is moved or changed, and exact copies are skipped.
⌘Z moves the pictures that have just come in to Recently deleted.
Stop on Generate stops watching and clears the folder setting.
Coming later
Making the pictures from Skye Desk, with ComfyUI on your Mac or on a rented GPU.
Shots, prompts and resultsUse your LoRA
Keep the prompts and the pictures you make with your LoRA in order.
Generate watches your ComfyUI output folder: new pictures appear by themselves, grouped by shot, ready to review.
Coming later
Making the pictures from Skye Desk with the Generate button.
Find the workflow in a pictureUse your LoRA
Pictures made in ComfyUI carry their workflow. Skye Desk reads it back.
Library and referencesUse your LoRA
Keep model files, your designs and reference pictures next to your projects.
Videos work in the library too: each one shows a thumbnail that plays when you point at it, and opens full screen.
Coming later
Downloading models from Hugging Face or Civitai.
Keyboard shortcutsReference
The keys worth learning first.
The ones to learn first
Privacy mode
Ten languages
Click the keyboard, then press ⌘K, Space or C.
Opens the first page in the top bar; ⌃2 to ⌃8 open the others, left to right.
Finds any page, project or command.
Searches the page you are on.
Previews the selected photo full size.
Moves to the previous photo.
Moves to the next photo.
Renames the selected photos.
Moves the selected photos to Recently deleted.
Undoes the last change, even a batch.
Compares the selected pictures side by side.
Resets the zoom.
In Runs, sets the final version.
In Generate, passes a result.
In Generate, rejects a result.
Opens the Filters bar.
Opens the number rules: tag count, words, tokens.
Opens the export panel.
Saves the caption you are writing.
Turns privacy mode on: every picture is blurred while you share your screen.
Pastes a screenshot into a reference board.
In a blind test, votes for neither.
Not a Skye Desk shortcut.
⌃1–⌃8 open the pages in the top bar, left to right.
⌘K finds any page, project or command.
⌘F searches the page you are on.
Space previews the selected photo; ← and → move between photos.
Return renames; ⌘⌫ moves to Recently deleted; ⌘Z undoes.
C compares the selected pictures side by side; 0 resets the zoom.
In Runs, F sets the final version. In Generate, A passes and X rejects.
⇧⌘H turns privacy mode on and off.
The Help menu lists every shortcut.
Press ⇧⌘H before you share your screen: every picture is blurred, the status bar says privacy mode is on, and it stays on until you turn it off or quit.
Skye Desk speaks the same ten languages as this site: English, Simplified and Traditional Chinese, Japanese, Korean, Spanish, Brazilian Portuguese, French, German and Italian.
Words you will meetReference
Short meanings, in the order you meet them.
A small add-on model that learns one subject – a coat, a character, a product, a style – from your photos. With it, an AI image model draws that subject the same way every time.
The big AI image model your LoRA adds to. A LoRA works with the model family it was trained on.
A made-up word, such as my_coat, that calls up your subject in a prompt.
The sentence that describes one photo. The AI learns from each photo and its caption together.
The photos and captions of one subject, kept together in one project.
One round of learning. Training saves a copy every so many steps, so you can compare them.
A saved copy of the LoRA at one step. The final version is the checkpoint you keep.
The text you give an image model to describe the picture you want.
Privacy and consentReference
What leaves your Mac, and whose photos you may use.
Your photos stay on your Mac. Nothing is uploaded unless you start a cloud run.
Train on a real person only with their written consent, and never on anyone under 18.
Do not train on other people’s work or likeness without permission.
Privacy policy
Can I use it commercially?Reference
It depends on the base model’s licence and on your right to use the photos.
Base models
Your photos
Models that draft captions
The pictures, captions and LoRAs you make are yours. Whether you may sell them, or make them for a client, depends on two things: the licence of the base model you trained on, and your right to use the photos you trained with.
On Launch, the line under the base model names its licence, and the Pre-flight check warns you when a base model does not allow commercial use.
AllowedThe licence allows commercial use. Some ask you to keep their notice.
Read its licence firstThe licence has its own rules for commercial use. Read them before you sell anything.
Not allowedThe licence is for non-commercial use. Do not sell the LoRA or paid work made with it.
This is a short guide, not legal advice, and licences change. The licence text of each model is what counts: read it before you sell anything.
Train on photos you took yourself, or photos you have permission to use for training.
Train on a real person only with their written consent, and never on anyone under 18.
Logos, brands and characters that belong to someone else can be protected even when the base model allows commercial use. Leave them out, or get permission.
Stock photos and pictures from the web have their own licences. Check that they allow training.
The AI models on your Mac that draft captions have licences too. When they arrive, Skye Desk will only offer models that allow commercial use, some with conditions.
AI models on your Mac
When something goes wrongReference
Quick fixes for the usual snags.
Open System Settings › Privacy & Security and click Open Anyway.
The import preview counts what is new, what is already there and any problems. Skye Desk reads PNG, JPG, WebP and BMP, and turns HEIC and TIFF into PNG. Design files, RAW photos and documents are left out: export them as PNG or JPG first.
They come in as PNG copies, and the import preview counts them under Converted to PNG. The photos you started from stay as they are.
In Finder, choose Download Now for those photos, then import them again.
A ZIP can hold up to 25,000 entries, 2 GB unpacked and 256 MB per file. Split a bigger one. For RAR or 7z, unzip it in Finder first and import the folder.
The caption file needs the same name as its photo: coat_01.jpg and coat_01.txt.
Every export is a full copy of its photos. Delete the old exports you no longer need, then empty the Trash.
On Launch, choose the GPU. Skye Desk lists what to change so that the run fits – quantising, low-VRAM mode, smaller pictures, batch 1 – and applies it in one click.
Point it at the folder HOW TO USE.txt names: the img folder for kohya_ss, the folder named after your trigger word for OneTrainer, the dataset folder for AI Toolkit. Each export is laid out for one trainer, so export again if you picked another one.
Check the ComfyUI output folder in Settings › Generate, and keep your project open on Generate. Only pictures made after you set the folder come in.
Shoot three to five different backgrounds, or describe the background in each caption. Tags shows when one background takes over.
Coming later Click Resume: it carries on where it stopped. A model is only used once every file matches Hugging Face’s checksum.
Coming later Cloud training is not in the first beta. Until then, export the dataset and train with your own tool.
Read the FAQ
- LoRA
- A small add-on model that learns one subject – a coat, a character, a product, a style – from your photos. With it, an AI image model draws that subject the same way every time.
- Base model
- The big AI image model your LoRA adds to. A LoRA works with the model family it was trained on.
- Trigger word
- A made-up word, such as my_coat, that calls up your subject in a prompt.
- Caption
- The sentence that describes one photo. The AI learns from each photo and its caption together.
- Dataset
- The photos and captions of one subject, kept together in one project.
- Step
- One round of learning. Training saves a copy every so many steps, so you can compare them.
- Checkpoint
- A saved copy of the LoRA at one step. The final version is the checkpoint you keep.
- Prompt
- The text you give an image model to describe the picture you want.
- ⌃1Opens the first page in the top bar; ⌃2 to ⌃8 open the others, left to right.
- ⌘KFinds any page, project or command.
- ⌘FSearches the page you are on.
- SpacePreviews the selected photo full size.
- ←Moves to the previous photo.
- →Moves to the next photo.
- ReturnRenames the selected photos.
- ⌘⌫Moves the selected photos to Recently deleted.
- ⌘ZUndoes the last change, even a batch.
- CCompares the selected pictures side by side.
- 0Resets the zoom.
- FIn Runs, sets the final version.
- AIn Generate, passes a result.
- XIn Generate, rejects a result.
- \Opens the Filters bar.
- ⇧⌘FOpens the number rules: tag count, words, tokens.
- ⌥⌘EOpens the export panel.
- ⌘SSaves the caption you are writing.
- ⇧⌘HTurns privacy mode on: every picture is blurred while you share your screen.
- ⌘VPastes a screenshot into a reference board.
- ↓In a blind test, votes for neither.