Upload sharp, well-lit sources
Use photos with readable detail on the subject you want to keep—product labels, faces, or key props. Avoid extreme blur, heavy JPEG artifacts, or clipped highlights when possible.
YinatrixAI Tools / Image
Image to Image AI
Anchor generation to your own photos or renders. Upload references, write a keep/change prompt, and produce controlled variations—background swaps, style shifts, and campaign-ready product scenes.
- 1–8 reference images (at least one required)
- Nano Banana 2 and GPT Image 2
- Preserve subject identity while changing scene, style, or mood
Reference-anchored edits (not prompt-only)
Up to 8 images for combine & style guides
GPT Image 2: JPEG, PNG, WebP refs up to 10 MB
Models
Which models support Image to Image?
Open Dashboard → Image to Image. The model picker loads from the live catalog; these are the image models wired into this workflow today. In the future, we will select more AI image models with unique expertise.
- Nano Banana 2 — fast iteration on edits; widest aspect-ratio list; 1K / 2K / 4K quality
- GPT Image 2 — strong fidelity for hero assets; JPEG, PNG, or WebP references only (10 MB each); adds ratios including 2:1, 1:2, and Auto
Reference uploads
What makes a good reference image?
The model anchors to your uploads. Sharp, well-lit sources with clear subjects produce fewer retries than blurry or heavily compressed files.
Assign a role to each reference
With multiple uploads, name roles in the prompt: “Use figure and pose from image 1, color palette from image 2, environment from image 3.” The UI supports up to eight references for combine workflows.
Prompting
Image to Image prompting: preserve first, then change
Lead with preservation clauses, then list changes. Smaller, specific edits succeed more often than “make it completely different.”
Background swap
Example: “Preserve the subject, pose, and outfit exactly. Replace background with a soft sunset beach; match warm key light on the subject.” Keep foreground and background lighting consistent.
Style transfer
Example: “Keep composition and subject identity. Re-render as watercolor illustration with soft pigment edges—no photoreal skin.” Avoid mixing conflicting style words in one prompt.
When to use this tool
Image to Image vs other tools — where it fits
Image to Image is the “reference still → edited still” path. It reduces randomness by anchoring to pixels you already approve.
- Choose Image to Image when: you have a photo, render, or Text-to-Image winner and need controlled edits—not a blank-slate concept.
- Choose Text to Image when: you are exploring directions without a reference file yet.
- Choose Image to Video when: you have a first frame (and optional last frame) and need motion between compositions.
- Choose Multi-images to Video when: you have several still guides (character, product angles, style board) for a Seedance 2.0 shot—not a defined end frame.
- Common pipeline: Text to Image → pick winner → Image to Image for background, style, or campaign variants → optional Image to Video for motion.
Transformations
What you can change while keeping the anchor
High-intent edits teams run every week—all from references plus a focused prompt.
Background swap
Keep the subject and foreground; replace the scene and match lighting.
Style transfer
Realistic ↔ illustration ↔ cinematic while preserving structure.
Lighting & mood
Day/night, soft/hard, warm/cool—dial the feeling without recomposing.
Composition refinement
Tighten crop, adjust framing, or focus without a full restart.
Product scene variations
Same product, different lifestyle or studio settings for campaigns.
Brand alignment
Match palette, tone, and art direction to your campaign system.
Prompt blueprint
Copy this keep / change structure
- Preserve: subject identity, pose, product label, layout, or brand marks
- Change: background, props, wardrobe, color grade, or art medium
- Lighting: describe key light direction so the subject matches the new scene
- Scope: one major change per generation when results drift
- Multi-ref: “Image 1 = subject, image 2 = palette, image 3 = environment”
- Avoid: vague “make it better” with no visual constraints
Examples
Example Image to Image outputs
Swap in your best before/after edits to build trust from organic search.
Portrait background edit
Subject preserved; environment and bokeh replaced.
Interior scene change
Product or furniture anchored; room style updated.
Fashion vertical ad
Outfit kept; dramatic location swap for 9:16 social.
Multi-reference merge
Elements combined from several still guides into one cohesive frame.
FAQ
What is image to image AI?
It uses one or more reference uploads as an anchor, then generates a new still from your prompt. You control what stays and what changes—ideal for edits, not blank-slate ideation.
How many reference images can I upload?
At least one is required; you can add up to eight. Generate stays disabled until at least one upload completes.
Which models work with Image to Image?
Nano Banana 2 and GPT Image 2 (from the live catalog). GPT Image 2 only accepts JPEG, PNG, or WebP references up to 10 MB each.
How long can my prompt be?
Up to 4,000 characters with a soft reminder above 1,000. Lead with preservation clauses, then list changes.
How do I change only the background?
Tell the model to preserve the subject and foreground, then describe the new background and matching light direction.
When should I use Text to Image instead?
Use Text to Image for exploration when you have no reference yet. Move to Image to Image once you have a base worth refining.
Why did GPT Image 2 reject my upload?
GPT Image 2 references must be JPEG, PNG, or WebP and no larger than 10 MB per file. Other models allow up to 12 MB per image in the default uploader.
Can I use outputs commercially?
Commercial usage depends on your plan and terms. Review the pricing page and policies before shipping at scale.
Edit from references—not from scratch
Upload, describe keep/change, and ship production stills without losing your anchor.
Related guides and tools
Text to Image
Explore concepts from a prompt, then bring the winner here for controlled edits.
Image to Video
Animate a strong still with first-frame (and optional last-frame) control.
GPT Image 2 (AI Power)
High-fidelity references and final hero assets with strict format rules.
Nano Banana 2 (AI Power)
Fast iteration on edits and the widest aspect-ratio list.
Pricing
Compare plans and credit packs for consistent production volume.