Keep Two AI Influencers Distinct in One Image
Your AI influencer looks recognizable alone. Add a second person, and the scene becomes harder to judge: the outfits are right, the pose looks natural, but the companion seems to have borrowed the main influencer's face.
A shared image can give you a different kind of post—a conversation, two ways to wear a color, or two fictional viewpoints on the same style choice. For that image to work, you need to recognize each person independently. Two different shirts are not enough.
The useful starting point is a simple scene brief that gives each person their own reference, appearance and place in the picture. Then you can compare the result with those references and ask for a specific correction. Here is how to work through that process without redesigning either AI influencer.
Give each reference one clear role
Start with two identities you have already chosen. Use images you are authorized to upload, with each person's face clearly visible. If you are still deciding what either person looks like, finish that AI influencer reference sheet first; a shared scene is a difficult place to settle two unfinished identities.
You also need an image tool that accepts multiple reference images for a single generation. For example, Google's Gemini image-generation documentation describes multiple-image input and assigning elements from different references to a new scene. The controls available in a particular app may differ, so check that you can supply both identities before adapting the prompt below.
For this illustrative example, imagine two fictional adult style influencers, Nora and Imani, seated together to show a rust-and-cream color pairing. Their selected references establish these appearances:
| Person | Identity details to preserve | Clothing for this scene |
|---|---|---|
| Nora | Long oval face, narrow jaw, straight dark chin-length bob, light olive complexion | Rust overshirt |
| Imani | Rounder face, broad jaw, short tightly curled black hair, deep brown complexion | Cream knit top |
These descriptions should match the reference images. They are reminders of what belongs to each person, not instructions to make established identities more different. For your own pair, describe the faces you actually chose.
Keep the clothing separate in your notes because an outfit can change while the person stays recognizable. Likewise, “on the left” is a position for this picture, not part of Nora's identity.
Attach Nora's chosen reference first and Imani's second if your tool exposes their order. In your prompt, call them “the first attached image” and “the second attached image.” If the app uses another way to identify references, use that instead. Don't assume a private filename or a label in your notes is visible to the generator.
Make the first scene easy to inspect
For Nora and Imani, start with a seated image framed from head to knees, both faces turned mostly toward the camera. Keep the pair large enough in the frame to compare their facial features. Leave a small gap between their shoulders and give each person a separate hand position. This makes it easier to see whose face and body you are reviewing than a hug, overlapping arms or a close profile conversation would.
The image can still have a purpose. Here, the rust overshirt and cream knit show a color pairing; a relaxed glance and slight turn of the shoulders suggest conversation. You can explore a more involved interaction after you have a candidate in which both identities look right.
With the two references attached in the stated order, try a brief like this:
Create one realistic image of two fictional adults seated on separate chairs beside a plain café wall. Use the person in the first attached image for Nora, on the viewer's left. Preserve her long oval face, narrow jaw, light olive complexion and straight dark chin-length bob. She wears a rust overshirt, with both hands resting separately on her lap.
Use the person in the second attached image for Imani, on the viewer's right. Preserve her rounder face, broad jaw, deep brown complexion and short tightly curled black hair. She wears a cream knit top, with her hands resting on her own knees.
Frame both people from head to knees at eye level, keeping their faces large enough to inspect, with a small gap between their shoulders. Keep both faces mostly toward the camera, with relaxed expressions and a slight turn toward each other. Use soft daylight and a simple background. Keep each person's facial features, complexion, hair and visible body proportions matched to her own reference.
Replace the example's identity details with yours. Avoid leaving its names or appearance descriptions in a prompt for a different pair; the text and attached images should tell the same story.
Check the people before judging the picture
Open the result beside both original references. Compare Nora with Nora first, then Imani with Imani. Look at face shape, jaw, hairline and hairstyle, complexion, and visible shoulder and torso proportions. Allow for the changed expression and lighting, but investigate features that now seem to belong to the other person.
Don't let a successful outfit distract you. Suppose the candidate puts a rust overshirt on the left and a cream knit on the right, just as requested. Nora still resembles her reference, but Imani now has Nora's straight bob and narrow jaw. The layout passed; Imani's identity did not. Rewording the cream top or swapping the chairs would miss the problem.
There is a different decision if both faces match their own references but Nora and Imani have simply exchanged sides. That is a placement issue. You can keep the image if the arrangement still serves the post, or request the intended positions if they matter. There is no reason to rewrite their facial descriptions solely because the people changed places.
Finally, check the scene around them: each arm should belong to the right body, hands should be plausible, and the chairs should support the seated poses. A recognizable pair can still be in an unusable picture.
Revise the mismatch you can name
For the hypothetical candidate where Imani acquired Nora's hair and jaw, keep the original references. Don't replace Imani's reference with the failed result; that would give your next attempt the very face you are trying to correct.
If your tool supports editing an existing image with reference inputs, identify the candidate as the scene to edit and the two originals as identity references. Ask for the specific change:
In the shared scene, correct Imani, the person on the viewer's right in the cream knit top. Match her face to her original identity reference: a rounder face, broad jaw, deep brown complexion and short tightly curled black hair. Nora's straight bob and narrow jaw belong only to Nora, on the viewer's left. Keep Nora matched to her own original reference. Retain the seated composition, clothing and daylight.
Use the tool's actual reference labels or attachment order in place of “her original identity reference.” The edit request describes what you want preserved; inspect the returned image to see what actually stayed the same.
If your tool cannot use those references while editing, make a fresh attempt with the original pair and the same simple scene brief. Strengthen the mapping in Imani's paragraph: “Imani's face and hair come from the second attached image only; Nora's appearance comes from the first.” Leave the outfits, framing and background alone for this attempt so the correction remains easy to assess.
Now compare both people again. Imani getting her curls back is not a complete fix if her jaw still looks like Nora's—or if Nora's face changed in the process. Keep the candidate only when each person is recognizable against her own reference and the shared scene is usable.
Build the interaction from a pair you recognize
For Nora and Imani's first color-pairing post, a simple seated picture can be enough. You do not need to add a hug, a shared prop or a dramatic camera angle to justify showing them together. Describe the post as a fictional styling scene, rather than presenting it as evidence of a real meeting or friendship.
For your next attempt, choose the two references, write one identity paragraph for each person, and try one uncomplicated composition. If the identities remain unstable after a focused revision, return to separate portraits or a simpler shared pose. Once both people hold up to comparison, you have a useful starting point for adding more interaction—without accepting a new face just because the overall picture looks good.