Guides

    Nano Banana 2: change one locked still (8cr)

    Nano Banana 2 is an edit. Point at the object. A new text-to-image throws away the still you approved.

    Versely Team5 min read

    Nano Banana 2 is an edit. Point at the object. A new text-to-image throws away the still you approved.

    Nano Banana 2 is Google's ultra-fast lightweight image row for rapid creative iteration. The catalog lists three categories: text-to-image, image-to-image, and edit-image. This page is the third. Eight credits is the call. The model does not take a duration. audio is null. You are changing a picture someone already signed, not rolling a cousin of it from a paragraph.

    Eight credits to keep the plate

    The still you approved is the asset. A fresh generate from text is a new sample: new hands, new label, new light, a face that is almost the one legal signed. That is a different lottery. Image-to-image and edit-image are how you keep the lottery you already won.

    Point at the object. Name the change. Leave the rest of the frame alone. "Make the cap matte black, same angle" is an edit. "A bottle on a wet bar, cinematic" is a generate. The second sentence will not remember which bottle you meant.

    The AI photo editor is the Versely surface for that instruction. Nano Banana 2 is the Google row underneath it. Google's other models sit on the same bill. This page is only the edit.

    Fourteen references, not a mood board

    max_reference on this row is 14. That is a cap, not a quota. One locked still of the product, one of the room, one of the face — then stop. Dumping fourteen near-duplicates of the same hero is how the model averages two lighting setups into a third object.

    requires_image is false, which is how people talk themselves into skipping the upload. You can run it as text-to-image. The claim of this page is that you should not, once a still exists. Text-to-image is the sketch. Edit is the note on the sketch. Mixing them on a plate the client already liked is how the plate becomes a cousin.

    Quality settings on the row are 1K, 2K, and 4K. Pick the tier the delivery needs. Do not upscale later because you edited at 1K "to save thinking." The cheap mistake is the wrong object, not the wrong pixel count.

    Ratios the still already has

    Aspect on Nano Banana 2 runs from 8:1 through 1:8, plus the usual 16:9, 9:16, 1:1, and auto. Match the still you are editing. Asking for 9:16 on a 16:9 pack shot is a crop dressed as an edit. If the delivery needs a different ratio, crop after the object is right, or generate a new composition on purpose — do not pretend a ratio change is a cap colour.

    The edit instruction should be one object, one attribute. Jacket colour. Label text. Background cleanup. Chain a second edit if the first landed. Two attributes in one prompt is how you get a third jacket and a new room.

    If there is no approved still, you are not on this page. Start on text to image until someone can point at a frame. Then come back. Lock the still before you spend on motion is the same rule one step later, when the next credit is video.

    When a new generate is still the right call

    An edit cannot invent a camera that was never in the file. New angle, new room, a different product that only exists as a sentence — that is text-to-image, then maybe a second edit. Nano Banana 2 will try if you ask. Trying is not the job.

    Do not use this row to "fix" a video frame you intend to animate. Edit the still. Approve it. Then pick an image-to-video row. Video credits are a terrible identity debugger, and eight image credits are cheaper than discovering the label is wrong at second four.

    The ranked list for this category is best image editing model. This page is only Nano Banana 2: Google, eight credits, 1K / 2K / 4K, up to fourteen references, edit the object, keep the plate.

    FAQ

    Can Nano Banana 2 generate from text with no image?

    Yes. requires_image is false, and text-to-image is a listed category. Once a still is approved, treat that as a reason not to. A new text-to-image is a new still. The edit is how you keep the one you have.

    Why is this 8 credits if it is "lightweight"?

    That is the catalog figure for the call. Discounted billing on the row exists in the snapshot; the job does not change with the discount. Lightweight is the generation style, not a licence to reroll identity.

    Do I need all 14 reference slots?

    No. Empty slots beat junk. One locked still and a short instruction outperform a collage. Fourteen is the ceiling, not a score.

    Should I edit at 4K every time?

    Only if delivery is 4K. The quality enum is 1K, 2K, and 4K. Pick the tier you will ship. Editing a social crop at 4K does not make the object more correct.