How to Remove Backgrounds From Images with Nano Banana 2

Start with the cleanest source image you can produce — a product on a plain background, a person against a contrasting backdrop, or a clean studio shot. The cleaner the source, the cleaner the cutout. For AI-generated source images on Astorie, generate at 2K or 4K (Nano Banana 2 supports both) so the cutout has high-resolution edge data to work with. For uploaded photography, ensure the subject is in focus and well-separated from the background; the background-removal model handles soft edges and color contrast, but a hand-blurred shot or a subject that matches the background color will exceed automated cutout quality.

Add a Tool node and select the workspace's background removal route (Bria RMBG is the typical default; see the `tools/background-removal` page for routing details). Connect the source Image node to the Tool input. The output is a transparent PNG with the subject isolated. For most subjects this is one-click — no prompts, no parameters. Edge quality on hair, fur, glass, and fine detail is the typical pain point: clean studio subjects come out near-perfect, while complex hair or transparent objects may need a second pass or manual touch-up. For AI compositing pipelines (the focus of this scenario), small edge imperfections often disappear in the next step when Nano Banana 2 re-renders the subject into a new scene.

Add a Nano Banana 2 Image node and pin the cutout PNG as one of up to 10 reference images. Optionally pin a second reference (a target lifestyle scene, a brand backdrop, or an aesthetic mood board) to direct the composite's look. Then prompt the composite explicitly: "the product from reference 1 placed on a marble countertop with morning light from the left, lifestyle photography style." Nano Banana 2 generates a new image with the subject preserved (face, product, identity) and the new environment rendered around it. This produces compositing results that beat traditional Photoshop cut-and-paste because the lighting, shadow, and color grading are unified across subject and environment in one render.

For static deliverables (PDP photography, ad creative, social hero images), output Nano Banana 2 at 4K and export. For video pipelines, feed the composited still into a video model (Seedance 2.0 for product motion, Sora 2 for narrative, Kling 3.0 for human performance) as the starting frame — the video model animates the composite scene while preserving the subject identity. This is the canonical "image cutout → AI lifestyle scene → video" pipeline that closes the gap between static product photography and motion content. The full chain runs on the Astorie canvas without external tools; templates can capture the exact node arrangement and re-run for the next SKU.