Removing the background from an image — local first, AI for tricky edges

Design Studio has a one-click background remover on every image you place on the canvas. Two modes — local (free) and AI refine (~1¢) — so you only pay for the edge cases that actually need it.

To use:
1. Click the image on the canvas to select it.
2. The right-rail Properties panel shows "Remove background" near the bottom of the image controls.
3. Click "Remove background" — the local model runs and swaps your image in place with a transparent-PNG version. Usually 2–10 seconds depending on size.
4. If the cutout looks great, you're done. If hair, fur, or translucent edges look rough, click the smaller "Edges look off? Refine with AI (~1¢)" link below.

Local mode (default — free, in-browser):
• Runs entirely on your device using the @imgly/background-removal WASM model
• ~30 MB model, lazy-loaded the first time you click — subsequent runs are instant
• No data leaves your browser; no chargeback to your org
• Works great on most photos with clear subject/background separation (products, people, vehicles)
• Trades off a bit on very fine edges (loose hair, fur, smoke, fluffy fabric)

AI refine mode (~1¢):
• Calls Replicate Bria-RMBG (a model tuned for tricky edges)
• Better on hair, fur, translucent edges, fabric details, complex backgrounds
• Result is a fresh transparent PNG that swaps into the canvas same as local mode
• Charges roll up to orgs/{orgId}/usage at +15% — the title text on the button shows the rough cost so you're never surprised
• Use sparingly — most images only need local mode

Both modes preserve your filters:
Brightness/contrast/saturation/blur applied before the background remove are reapplied to the new transparent image, so your color treatments survive the cutout.

Re-running:
You can re-run either mode on the same image — each run starts from the current canvas image, so you can stack edits (e.g. local-cutout → AI-refine if local missed an area).

Undo:
Cmd/Ctrl+Z undoes the cutout and brings back the original image. The history stack pushes a step on every successful run, so you can always step back.

If it fails:
• "Failed to load processed image" — usually a network blip on the AI mode; try again, or switch to local mode
• Local model takes too long — your image may be very large; resize on the canvas before running (smaller image = faster cutout)
• Edges still look off after AI mode — the source image may have such soft separation that no model can recover it cleanly. Try cropping tighter around the subject before running.