Google Multi-Reference Image Editor
Google's lightweight editor blends a base photo with up to nine more references — products, styles, people, palettes — from a single plain-language instruction, in seconds.
Your generated image will appear here.
Enter a prompt and click 'Generate' — or start from an example:
Examples
Drag the handle to compare each source and result, or open the viewer for a closer look.
photo editing, identity lock
Prompt: Move this woman into a sunlit park for a relaxed outdoor headshot: soft golden-hour backlight, blurred green foliage behind her, a gentle breeze in her hair. Keep her face, hairstyle, blazer, white shirt, and warm confident smile exactly the same so she is unmistakably the same person.
product edits
Prompt: Restyle this skincare ad as a cozy winter-holiday variant: replace the eucalyptus leaves with frosted pine sprigs, scatter a few tiny warm fairy lights at the base, and shift the pale sage background to a deep forest green. Keep the bottle, gold cap, travertine pedestal, the headline "DEWY, NOT GREASY", and the label text exactly the same and perfectly legible.
photo editing
Prompt: Remove the two dark metal bollards from the foreground and any litter or stains on the pavement, rebuilding the cobblestones where they stood. Keep the café façade, the rattan chairs, the signage, the lighting, and the colors exactly the same.
Source: decar66 — CC-BY 2.0 via Flickr.
style changes
Prompt: Restyle this overhead lunch photo as a modern flat-lay illustration: clean vector-like shapes, soft grain texture, a warm cream background, and a terracotta, olive, and mustard palette. Keep the same composition — toast with burrata at the center, tomatoes and basil to the right, olive oil carafe to the left, napkin and fork at the edge.
product edits
Prompt: Place the mug on a bright café table by a large window: soft daylight, a blurred espresso bar in the background, a faint reflection on the tabletop, and a natural contact shadow under the mug. Keep the mug itself — color, handle position, lettering, small imperfections — exactly as in the original photo.
Source: RandalWhitmore — CC-BY 2.0 via Flickr.
style changes + character consistency
Prompt: Extend this robot mascot sticker sheet with a third row of three new stickers in the same style: the robot waving goodbye, the robot reading a tiny book, and the robot jumping with joy. Keep the same round white robot design, orange eye, leaf antenna, thick white sticker outlines, soft flat 3D style, and mint background perfectly consistent.
Nano Banana Edit is Google's lightweight image editing model, and its defining skill is combination: it accepts up to ten input images at once and weaves what it finds in them into a single coherent result. A base photo supplies the scene, and the other slots contribute whatever you point at — the jacket from image two, the lighting from image three, the palette from a moodboard frame.
On RenderFlow AI, the image editor canvas exposes all ten reference slots plus a PNG/JPEG export choice at 5 credits per image. The model answers in seconds, which makes it the fastest way on the platform to test whether two visual ideas actually work together before committing to a slower premium edit.
Upload the photo that should define the composition — the person, the room, the scene — as the first input. Nano Banana Edit treats it as the base that the other references modify, so choose the frame whose layout you want to keep.
Enable Inputs 2–10 for the elements you are pulling in: a product shot, a garment, a style frame, a palette. Fewer, clearer references beat a full rack — every extra image is another thing the model can confuse, so toggle on slots deliberately.
Write the instruction with explicit assignments: "put the jacket from image 2 on the person in image 1, lit like image 3." Role-naming is the single biggest lever for getting the right element from the right reference.
The canvas offers both formats. Take PNG when the result feeds further design work or needs maximum fidelity; take JPEG for quick social posting where file size matters more.
Keep a good output as the new base image and issue the next instruction to build on it. When a job only involves one photo with no references to blend, Qwen Image Edit at 3 credits is the cheaper tool for that pass.
Up to ten inputs on the RenderFlow AI canvas: one required base image plus nine optional references. Give each reference a distinct job — garment, background, style, palette — and the model blends them into a single coherent result.
Nano Banana generates images from a text prompt alone, while Nano Banana Edit starts from your uploaded images and changes or combines them. If you have source material to preserve or remix, you want the Edit page; if you are creating from scratch, use the generator.
Yes — character consistency is one of its strengths. Provide a clear portrait as a reference, describe the new scene or styling, and the person stays recognizable; reusing the output as the next base image tightens consistency further across a series.
Both are offered in the canvas. PNG preserves maximum detail for assets headed into design tools or further edits; JPEG produces smaller files that are fine for social posts and quick reviews.
The usual cause is too many similar images or an instruction that never says which image does what. Cut the reference count to what the edit truly needs and rewrite the prompt with explicit roles — "background from image 2, product from image 3" — and the confusion almost always clears.
It is one of its best uses: upload the packshot as one input and a lifestyle scene as another, then ask for the product placed in the scene with matching light. Review labels and fine print closely, and rerun with a sharper packshot if small packaging text softens.
Generate images with Google Nano Banana.
Edit images with ByteDance Seedream V4 Edit.
Use the premium FLUX Kontext image editing model.
Edit images with Alibaba Qwen prompt-based controls.
Use advanced Alibaba Qwen image editing for product and layout changes.
Edit photos with ByteDance SeedEdit V3 prompt-based image editing.