AI Image Combiner

How to combine two images

1. Prepare the instructions above and open the studio. This selects Nano Banana 2.
2. Upload your subject photo in the first field and your base scene in the second. Use JPG, PNG or WebP, up to 5 MB each.
3. Review the order, prompt, aspect ratio and 6-credit cost, then Generate once. Inspect and download the result from the existing result panel.

The order matters. Reference 1 identifies the object to bring into the scene. Reference 2 supplies the setting. Describe placement using something visible in the base scene, such as the table on the left or the empty pedestal in the center. Avoid referring to objects that are not in either input. A precise request is easier to evaluate than a list of unrelated visual ideas.

What did our two-image test show?

We tested an existing synthetic product image against three simple backgrounds drawn for the experiment. Eight API runs produced images. The bottle and its label remained recognizable in the reviewed outputs, but some results added objects, changed the pedestal or carried details from the original background into the new scene. The initial prompt also contained an irrelevant instruction about another scene, so those first results are not a fair measure of general model reliability.

After correcting the prompt, neither of the two additional outputs added a plant. One round-pedestal composition was acceptable for that controlled static example. The other still brought source-image water drops into the new scene and changed the pedestal. This is a small feasibility check using synthetic material, not evidence of reliable person insertion, face preservation, transparent-object compositing or unchanged background pixels.

Use this for a new composition, and review the details

A generative combiner creates a new image guided by two references. It does not simply paste untouched source pixels on top of one another. Product proportions, lettering, reflections, shadows and background objects may change. If an exact logo, label or photographed detail matters, compare the result against the original at full size. Keep the source files and reject a version that misrepresents the object.

Choose clear input images with one main subject and a base scene that has room for it. Matching viewpoint and lighting can make your requested arrangement easier to describe. If the source has a large headline or a second object, explicitly state whether it belongs in the result. Do not ask the model to preserve an entire source image while simultaneously replacing its setting. State the elements to retain and the changes to make separately.

Common questions

Is AI image combining free?

This workflow uses a paid image-generation model and costs 6 credits per submitted generation. Your available balance and cost appear in the studio. The local Stitch and Photo Collage tools are free because they arrange existing pixels in your browser without a model call. Choose one of those when a side-by-side layout is enough.

What happens if generation fails?

The workflow uses the same job and credit system as ordinary image generation. Credits are reserved for the job and failed or blocked generations follow the existing refund path. A completed image that does not match your preference is different from a technical failure. A new attempt is another generation and uses credits. Review the current status before submitting again.

Are my photos kept in the browser?

The selected references are uploaded when you submit a generation and are processed by the image provider. This is different from the local editing tools. Browser drafts keep prompt settings, not the reference image bytes. After refreshing, re-upload the required images before making a new request. Your completed result uses the same account history and private download path as the other image models.

Can I insert a person without changing their face?

We have not established that capability with these product tests. Two-reference support alone does not prove reliable face identity, pose, person count or occlusion. Do not use this page as a promise of precise person insertion. Review any identity-sensitive result carefully and use images you are authorized to process.

Does this generate video?

No. The output is one still image. This workflow does not animate either reference, generate audio or create a video clip. GIF Maker can arrange already-existing stills into an animation, but it does not generate motion between them.

Related workflows

Read the tested workflow · Stitching vs AI combining · Arrange photos without AI · Compare your result · Add exact text · Read the model guide