v1.3.4API credit, right in the headerChangelog →
MergeRef
Notify Me
Tools

Object Insert (Gemini)

Put an object from another image into a scene: a product on a table, a bag on a shoulder, a character in a background. Object Insert runs on Gemini image models and appears in the prompt panel when one is selected.

Object Insert — demo video

Place it, and the AI works out the rest

Add the object image (upload it, or pick it from the canvas), then move, scale and rotate it where it should go. Pasting is the easy part. Making it belong is the hard part: the right size for its depth, feet on the ground, the scene's light on it.

AI placement analysis

Every time you generate, the placement analysis looks at the scene with your objects in place and describes each one in words for the image model:

  • What it is, specific enough to keep its identity. For a person: apparent age, hair, clothing.
  • Where it stands, against what is really in the scene: people, furniture, ground, walls, foreground or background.
  • How big it should read next to nearby people and things, and whether that fits its depth.
  • How it meets the scene: the surface it rests on, what passes in front of it, the light and shadow it should catch.
  • What to complete when the object is cut off and the scene needs the rest, such as legs down to the sidewalk.
  • The scene, once: light (direction, quality, colour temperature), style (photo or illustration, grain, depth of field, colour grade) and camera (eye level, lens feel).
  • Your prompt steers it. What you type goes to the analysis as the request, so it reads placement for that intent. With no prompt it judges from the scene alone.

It sees the scene with each object outlined and numbered, the scene before pasting (so it knows what the objects now cover), and each object on its own. Positions and sizes go in as words, never coordinates, because the image model follows descriptions better. If the analysis is unavailable, only the rough position and size are sent.

Assist AIModelAbout per generate
GeminiGemini 3.8 Flash≈ $0.01
ChatGPTGPT-6.1 Sol≈ $0.02

For one object; a little more for each extra object. Same official prices as Inpainting (checked 4 November 2026, Google · OpenAI), billed to your own key. One Nano Banana 2.1 image at 1K, the default model, is about $0.034.

Compositing presets

A preset tells the model what kind of composite this is, so it knows what to keep and what to adapt. The analysis also suggests one (marked AI, never applied for you). There are 26 presets in six groups:

  • Basic: Place · Hold · Wear · Add person
  • Commerce: Studio product · Lifestyle · Keep packaging (text, logo, colours and proportions unchanged; only light, shadow and perspective adapt)
  • Fashion: Accessory · Outfit swap · Hair accessory
  • Interior: Furniture (follows the room's vanishing lines) · Wall mount · Exterior
  • Advertising: Hero product · Crowd extra · Vehicle
  • Illustration: Weapon · gear · Costume part · Character into BG · Mob character · BG object · Effect · Item · prop · Photo → art · 3D → 2D · Line art · Key art · Merch mockup