1
Upload a strong reference image
Start with a clean portrait, product photo, or source frame so the model has clear identity, structure, and visual cues to preserve.
Back to all features
Reference-led image editing
Turn one image into cleaner, sharper, or deliberately restyled variations
Image-to-image workflows are for the moments when the starting frame matters. Bring in a reference image, keep the structure you need, and use prompts to restyle, refine, or shift the result into a different visual direction.
ReferenceGenerated scene
One reference, multiple new scenes
One reference image can open multiple new scenes without losing the subject.
These examples start from one portrait reference and show how image-to-image can preserve identity while moving into professional branding, nighttime fashion, and luxury lifestyle directions.
A practical image-to-image workflow
The best image-to-image results come from a strong reference, clear preservation intent, and a prompt that describes the new scene you want to generate.
2
Decide what should stay the same
Be explicit about what must survive the generation, whether that is the face, subject proportions, product shape, or other core visual traits.
3
Prompt the new scene and review variations
Describe the environment, wardrobe, styling, lighting, or mood you want, then compare the variations that preserve the subject while opening a new direction.
Model guidance
Model choice should reflect how much structure you need to preserve.
Image-to-image is less about raw generation variety and more about matching the model to the kind of change you want to make.
Best for minimal drift
PreserveSeedream 5.0 Lite is a strong first choice when you need low-cost image-to-image work with solid preservation, especially for human portraits. If you need broader reliability across more subjects and are willing to pay more, Nano Banana 2 is the safer general-purpose upgrade.
Best for cleanup and refinement
PolishFor most cleanup and refinement work, Seedream 5.0 Lite delivers excellent results for the price. When the output still needs stronger control or a more dependable finish, Nano Banana 2 is the model to reach for.
Best for stronger reinterpretation
RestyleNano Banana 2 and Nano Banana Pro are the strongest choices when the goal is a bigger shift in mood, environment, styling, or scene construction. They are the most capable and flexible models on this page for heavier reinterpretation.
Image-to-image is strongest when the reference already matters.
Use this workflow when the subject, face, product, or source frame should anchor the result, even if the final scene looks very different.
Best when you need
New scenes from an existing subjectIdentity continuity from a referenceControlled wardrobe or background changesReference-guided restyles and refinement
Use another workflow when you need
Prompt-only ideation from scratchOutputs with no reference continuityMotion from a still frameLarge coordinated output runs
Connect this workflow to the rest of your stack
These routes help you move from feature discovery to inspiration and model selection without losing the thread.
Feature
Batch Generation
Launch prompt variations, ratios, and reference sets in one batch so your team reviews the full output set together.
Explore
Feature
Text to Image
Turn prompts into polished images for campaigns, product visuals, portraits, and concept work.
Explore
Comparison
Model Comparisons
Compare models side by side before you commit a workflow.
Explore
Inspiration
Real estate headshots
Listing-ready styles for agents and teams.
Explore
Image-to-image FAQ
How much of the original image can I preserve?
That depends on the strength of the reference and how aggressive the requested change is. Image-to-image works best when the prompt targets specific changes instead of rewriting the entire scene.
What kind of reference image works best?
Use a clean, sharp image with readable subject detail. The stronger the source frame, the easier it is to preserve composition and identity where needed.
Can I edit the same image multiple times?
Yes. A common workflow is to keep the best variation from one pass, then use it as the starting point for the next refinement round.