AI Capabilities

LoongScales has built-in AI capabilities for image generation, background removal and image splitting. It appears at the bottom of the interface.

Open the AI panel

• There is a floating ✨ AI Generate pill button at the bottom of the interface.
• Click it, or press Ctrl+K (⌘K on Mac) from anywhere, and the AI panel rises from the bottom.
• The button can be dragged to the left/right edge of the canvas and auto-snaps; its position is remembered.
• The button has 4 states: Idle (blue breathing glow) / Generating (progress ring) / Done (red dot badge) / Error (red border).

LoongCoins & balance

AI features consume an in-app token called LoongCoins.

• The panel title bar always shows the current balance, e.g. 💰 1,234.
• The estimated cost is shown before each generation; when the balance is insufficient the button is disabled with a top-up prompt.
• The balance is deducted automatically after a successful generation.
• Click the 📜 Details button in the title bar to view LoongCoins details.

Three main tabs

The panel has three capabilities: Generate / Remove Background / Split Image. (Under the Generate tab you can pick a specific model.)

Generate (text-to-image / image-to-image)

The generate form includes:

• Prompt: describe the image you want in natural language.
• Reference layer: None (pure text-to-image) or Current selected layer (image-to-image).
• Output location:
  • New layer: result added as a new layer (default for text-to-image).
  • Replace current layer: result replaces the selected layer (auto-disabled when "no reference" is chosen).
• Size: a resolution tier (e.g. 1K/2K) + aspect ratio (e.g. 1:1, 16:9); the real pixel size shows live below.
• Quality tier (when the model supports it) and transparent background (when supported).
• Estimate / Generate button: shows cost and triggers generation.

Typical usage:

1. Text-to-image: reference=None, output=new layer, write a prompt, pick size, click generate → a new image.
2. Image-to-image (new): select a layer first, reference=current, output=new → AI generates a variant based on it, as a new layer.
3. Image-to-image (replace): reference=current, output=replace current → rewrites the selected layer directly (position/size preserved after transform).

AI generate panel: prompt box, reference layer, output location, size tier, generate button, LoongCoins balance.

Remove background

• Select a raster layer, then click the Remove Background tab in the context bar.
• One click cuts out the foreground and makes the background transparent; the result replaces the current layer (transform/position preserved).
• Background removal also costs LoongCoins and is recorded; it can be undone with one click (Ctrl+Z).
• If you don't want to remove the background, click the Remove Background tab again to return to the Generate screen.

Split image

• Select a raster layer, then click the Split Image tab in the context bar.
• AI splits one image into multiple independent parts and returns a layered PSD, which LoongScales imports automatically as multiple layers.
• Split Image uses the open-source framework See-Through: it decomposes a single image into fully inpainted, semantically independent layers and infers their drawing order — up to 23 layers, including hair, face, eyes, clothing, accessories, etc. It works best on cartoon portraits; other image types are not recommended.
• If you don't want to split, click the Split Image tab again to return to the Generate screen.

Generation history & "bring back what you deleted"

There is a History (🕘) button on the right of the panel title bar. Open it to see all generation history of the current work (cloud records) — every text-to-image, image-to-image, background removal and split.

Important: even if an AI-generated image is deleted from the layers, it can be brought back.

• In the history list, each record has a thumbnail and operation type;
• Click 「Add」 on that record to add that generated image back to the canvas (as a new layer);
• Split results (multiple layers) can also be added back — clicking "Add" on a split record re-imports it as multiple layers.

So you never worry about "I deleted the layer by accident after generating" — you can always pull the image / split result back from the generation history.

Not logged in / quota / errors

• If not logged in, clicking the AI button guides you to log in; the button only works after login.
• Insufficient balance: button disabled with a top-up prompt.
• Generation failed: the bottom button turns red and shows the reason (e.g. policy violation, timeout, network error); click once to return to normal.

Before/after of background removal + split image (combined → multiple sub-layers).