Image to Image AI: How to Edit Photos With a Prompt (2026 Guide)

Image to image AI lets you change a picture by describing the change. Upload a photo, type "turn it into a watercolor" or "put this bottle on a marble counter", and a model returns a new image that keeps your subject and follows your instruction. This guide covers how it works, how to get reliable results, and which model to pick for which job.

Quick answer

To edit an image with AI, open DeGemini's image to image tool, upload up to four reference images, describe what should change and what must stay the same, pick a model and click Generate. Results arrive in about 10 to 60 seconds, without a watermark, for 10 to 13 credits per image.

What is image to image AI?

Text to image models start from noise and a prompt. Image to image models start from a picture you provide. Early "img2img" tools simply re-noised your photo and redrew it, so you traded likeness for creativity with a strength slider. Today's editing models, such as Nano Banana 2 and GPT Image 2, read the reference image and the instruction together, which lets them change one thing (the background, the style, an outfit) while keeping the rest recognisably the same.

Typical uses:

• Restyling a photo as a painting, anime frame or 3D render
• Replacing or extending a background
• Turning a plain product photo into a studio or lifestyle shot
• Keeping a character consistent across several scenes
• Changing colours, clothing or the text on a sign
• Turning a sketch or wireframe into a finished picture

How to use image to image AI on DeGemini

1. Upload references. Add up to four PNG, JPG or WebP images by dropping, pasting or browsing. Large photos are resized to 2048 px in your browser before upload.
2. Describe the edit. Say what changes and what stays: "Keep the woman's face, hair and pose exactly the same. Replace the background with a rainy Tokyo street at night."
3. Pick a model and aspect ratio. On Nano Banana 2 and GPT Image 2, Auto keeps the shape of your first image. FLUX 2 Pro and Seedream 5 Pro preselect the closest standard ratio.
4. Generate and iterate. When the result appears, click Edit this image to use it as the next reference, so you can build an edit in small steps. Download when you are happy.

You need a free account so your credits and images stay yours. If a generation fails, its credits are refunded automatically.

Which model should you use?

All four models handle both text to image and image to image on DeGemini, at the same price in either mode.

Model	Made by	Best for	Credits
Nano Banana 2	Google	All-round edits that keep faces and products consistent; fast	13
GPT Image 2	OpenAI	Product shots, clean layouts, legible text inside the image	10
FLUX 2 Pro	Black Forest Labs	Polished, commercial-style graphics and illustration	10
Seedream 5 Pro	ByteDance	Photorealism and poster typography, including Chinese text	12

A simple rule: start with Nano Banana 2 when the subject must stay recognisable, switch to GPT Image 2 when the image contains words, use Seedream 5 Pro for realistic scenes or Chinese typography, and try FLUX 2 Pro for stylised marketing visuals.

Prompt formulas that work

Good edit prompts have three parts: what to keep, what to change, and how it should look.

• Restyle: "Keep the composition and the dog's pose. Repaint the whole image as a loose watercolor with soft paper texture."
• Background swap: "Keep the person, clothing and lighting on the face unchanged. Replace the background with a sunlit beach at golden hour, matching the light direction."
• Product shot: "Keep the bottle's shape, label and colours exactly. Place it on a white marble counter with soft morning window light and a shallow depth of field."
• Text edit: "Change the text on the shop sign to 'OPEN LATE' in the same font and colour. Change nothing else."
• Sketch to image: "Turn this pencil sketch into a finished digital illustration, same layout, bright flat colours."

Tips:

• Ask for one main change per run, then refine with Edit this image.
• Name the light and lens for realism: "overcast daylight, 35 mm, eye level".
• Say "change nothing else" when the edit should be surgical.

Working with several reference images

With more than one reference, tell the model which image provides what: "Use the woman from image 1 and the red jacket from image 2. Put her in the café from image 3." Keep the most important image first; on Nano Banana 2 and GPT Image 2, Auto follows the first image's shape. Up to four references are supported per edit.

Common problems and fixes

The face changed. Say explicitly which features must stay ("same face, same hairstyle, same expression"), and try Nano Banana 2.

The edit drifted too far. Shorten the prompt to one change, then iterate with Edit this image.

The text is garbled. Use GPT Image 2 or Seedream 5 Pro and put the exact words in quotes.

The framing is wrong. Pick the aspect ratio yourself instead of Auto, or crop your reference before uploading.

Pricing and privacy

Each image costs 10 to 13 credits depending on the model. Credit packs are one-time purchases: 100 credits for $2.99, 500 for $9.90 or 2,000 for $29.90, valid for 365 days. That works out to roughly $0.15 to $0.39 per image, with no subscription.

Reference images are sent to the AI provider only to run your edit and are deleted automatically within 24 hours. Only upload images you have the right to use, and never photos of other people without their consent. Results carry no visible watermark; some models embed invisible provenance data, such as Google's SynthID, that doesn't change how the image looks.

FAQ

Is there a free image to image AI? Many tools have free tiers with daily limits or watermarks. DeGemini charges per image instead, from 10 credits, with no watermark and no subscription. Our Gemini watermark remover stays free.

What's the difference between image to image and text to image? Image to image starts from your picture and edits it; text to image creates a new picture from words alone. Use the AI image generator for the second.

Can I use several photos at once? Yes, up to four reference images per edit.

How long does an edit take? Usually 10 to 60 seconds.

Ready to try it? Open image to image AI, upload a photo and describe the change.