Keep the person. Change the scene.
Give each reference a clear role so a new outfit or location does not turn into a new face.
The scene changes. The person should still be them.
That is the real challenge in a reference-led image. A photo contains a face, clothes, lighting, and a background; your prompt needs to say which of those you want to carry forward.
Choose one identity anchor
Open Image with an edit-capable model, or prepare the character in AI Influencer. Use a clear face reference with enough detail to recognize the person. Add a full-body view when the outfit or silhouette matters.
Keep that source stable while changing the scene. Several contradictory hairstyles or ages make a weaker reference pack, not a stronger one.
Give every image a job
Attach only what this edit needs, in an order you can explain:
Image 1 defines the adult character's face, hairstyle, and proportions. Image 2 defines only the cream jacket: its collar, buttons, and fabric. Photograph the character from Image 1 wearing that jacket in a quiet train station. Waist-up framing, soft daylight, unobstructed face. Do not borrow the jacket model's identity or background.
Use only as many images as the selected model supports. Check the attachment order before running.
| Reference | Keep from it |
|---|---|
| Identity | Face, proportions, distinguishing features |
| Outfit | Garment construction, color, fabric |
| Composition | Position, crop, viewpoint |
| Style | Light, palette, visual treatment |
Split a difficult edit into two wins
If changing the face reference, outfit, pose, and location together creates drift, simplify. First get the character wearing the jacket in a plain setting. Then move that approved version into the station.
Compare each step to the original identity: eyes, jaw, hairline, and distinctive details. If a later edit drifts, return to the clean reference rather than building on the mistake.
Try it: write one sentence for each attachment: “This image controls ___, but not ___.” If the roles overlap in a way you cannot resolve, remove one.
Your result is ready when the new scene is obvious and the same person is recognizable without an explanation.
Next: build a three-look fashion series.