Grok Imagine 2.0 Image to Image: Restyle and Revise Photos
Last verified: October 29, 2026
Grok Imagine 2.0 works best on pictures that are almost finished. You upload the version that is close, say what should change, and get back a revision that keeps what you brought. xAI built the model to follow instructions closely, down to the details, and to preserve what you put in across generations and edits. With Grok Imagine 2.0 Image to Image on Vidofy, you can also frame that revision as a full phone screen: 9:19.5 tall or 19.5:9 wide. xAI pairs those two ratios with modern iPhone displays, so a reworked photo can be framed for a lock screen or a vertical story from the first run.
An edit starts from a source image and a plain-language instruction. The model reads what is in the picture and applies the changes you ask for. xAI's launch notes say real work is iterative and the first generation is rarely the final asset. For a working creator, that means the image you already approved stays the anchor and the prompt does the revising. You don't need to rebuild the scene from nothing each time.
The form is short. Your uploads in Add Image set the subject. The Prompt carries the change. Aspect Ratio decides the frame, and Resolution Quality decides how much detail the returned image holds. Because every setting is visible before you generate, you can do quick draft passes and then a final render without leaving the page.
Explore more image to image
Inputs and output at a glance
What each run takes in and hands back.
Input
Up to 3 images (JPG, JPEG, PNG or WEBP, up to 10 MB each) + a text prompt of up to 5,000 characters
Resolution
1K or 2K
Aspect Ratio
Auto, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 1:2, 2:1, 9:19.5, 19.5:9, 9:20, 20:9
Output
1 image per run
How should you prompt Grok Imagine 2.0 for an edit?
Lead with the change, then lock the rest
Build the result in rounds instead of one long prompt
Describe the subject as something you could point at
How do you edit a photo with Grok Imagine 2.0?
Five steps, from upload to download.
Add your source pictures in "Add Image"
Start with the picture whose subject and layout you want to keep. Add more pictures only when the edit has to bring in a person, product or style from them. Use the sharpest original you have, because a soft or heavily compressed upload gives the edit less to work with. The accepted formats, file sizes and file count are in the FAQ.
Describe the change in the "Prompt" box
Write what should change first and what must stay the same after it. The box takes up to 5,000 characters, which is room for a full brief, but most edits need only a few precise sentences. If you only have a rough idea, the "Enhance Prompt" button under the box suggests fuller versions. Your text stays as written until you pick one.
Choose "Aspect Ratio" and "Resolution Quality"
"Aspect Ratio" offers Auto (the default), 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 1:2, 2:1, 9:19.5, 19.5:9, 9:20 and 20:9. Keep Auto while you test the change, and pick a fixed ratio once the image has a destination, so the frame is your decision. For a series, use the same ratio on every run: xAI's prompting guide recommends this so the images sit together, and suggests 3:2 for most photographs. "Resolution Quality" offers 1K (the default) or 2K. Draft at 1K and save 2K for the version you keep.
Run the edit with "Generate"
Click "Generate" once the uploads, prompt and settings are in place. Change only one thing between runs, whether that's a phrase, the ratio or an upload, so you can tell which choice caused which result.
Download the result
Before you download, compare the result with what your prompt said to keep: faces, labels, the horizon, the crop. If something drifted, note it. That note becomes the next round's instruction.
Editing the upload versus starting over from text
When an existing picture is worth editing, and when a new generation is the better start.
| Criterion | Grok Imagine 2.0 | Alternatives | Best For |
|---|---|---|---|
| You already have a picture that is close | An image-to-image edit fits when the upload already shows the right subject in the right layout and only a few details need to change. | Generating a new image from a text prompt fits when there is no picture yet to work from. | Edit when you have an approved photo or design to revise, and write a text prompt when you are still looking for the first version. |
| The subject itself is wrong | An edit is meant for changing details on a subject that is already right, with a prompt that holds the subject, angle and crop in place. | A new generation from text is the recommended route when the picture shows a different object, is missing a key element or has the wrong count. | Fix details with an edit. When the thing in the frame is wrong, rebuild the scene from text, then bring the new image back for finishing passes. |
Fixing common image-to-image problems
What usually goes wrong in an edit run, and what to change before the next one.
Add Image won't accept my file
The result came back a different shape than I wanted
A combined image has extra people, or someone is missing
Small details like faces in a crowd or fabric texture look soft
Frequently Asked Questions
How many images can I upload to Grok Imagine 2.0, and which files work?
Add Image takes one to three pictures per run. Files can be JPG, JPEG, PNG or WEBP, up to 10 MB each, and every picture must be at least 300×300 pixels. One upload is enough for a straight restyle or revision. Use the second and third slots for edits that bring in another person, product or style.
Why does one generation cost more than another?
Once you're signed in, the credit cost of a run is calculated from your choices and shown above the Generate button before you click it. Two things change it: the Resolution Quality you pick and the number of images you upload. Asking Enhance Prompt for suggestions doesn't use credits.
Which aspect ratio should I pick for a phone wallpaper or vertical story?
For a full-height iPhone screen, pick 9:19.5, or 19.5:9 for a landscape phone frame. xAI pairs those two with modern iPhone displays. It pairs the two ratios that follow them in the Aspect Ratio list with Android screens, and suggests the standard vertical and widescreen shapes for mobile and story formats. If one edit has to serve several placements, run it once per target ratio instead of leaving the frame on Auto.
Can I get several versions of an edit from one run?
No. Each run returns one image. To compare options, keep the same uploads and settings and change one phrase per run, so each result shows what that change did.