"Can I upload a reference photo and have an AI copy that exact pose onto my own face?" is a reasonable thing to want, and the honest answer for most AI photo generators — Pose AI included — is: not for still images, not yet. What you can do instead is choose a style or scene, or describe the pose and composition you want in text, and the identity-locked model renders your face into that. For video, the picture is different, and reference-based motion is genuinely more capable.
- For still photos, Pose AI does not currently support uploading a reference photo to copy its exact pose onto your face. You select a style or scene, or describe the pose you want, and identity-locked Nano Banana 2 generates you into it.
- For video, Pose's native SeedDance model accepts up to three video clips as a multimodal reference input, which means video generation can reference existing motion in a way still-image generation currently cannot.
- Nothing here is called "pose matching" as a branded feature — it's style-based pose selection for stills, and reference-based motion for video, and the two shouldn't be confused.
What reference-photo pose copying actually means
Reference-photo pose copying is the idea of uploading a photo of someone in a specific pose — arms crossed a certain way, a particular camera angle, a specific stance — and having an AI apply that exact pose to a different person's face and body. It's a real technique in some specialized computer-vision research (pose-transfer models that map a skeleton or keypoints from one image onto another subject), but it isn't a standard, widely shipped feature across mainstream AI photo generators, and it isn't one of Pose's current still-image features.
How Pose AI actually handles pose today
For still images, pose comes from the style or scene you select, or from a text description if you're using Pose's prompt-based generation — not from an uploaded reference photo. Identity-locked Nano Banana 2 then renders your face, read from your own selfie, into that described scene. If you want a specific stance or angle, describing it in words ("three-quarter angle, hand on hip, looking over the shoulder") is the current path, not uploading a second photo of someone else in that pose.
This is a real limitation worth stating plainly rather than glossing over: if your workflow depends on matching an exact reference pose for a still photo, Pose doesn't do that today.
Reference-pose support for still images
| Approach | Reference-photo pose copying? | How pose is actually set |
|---|---|---|
| Pose AI (stills) | No | Style/scene selection, or text description in prompt mode |
| General prompt-based generators | No, in most mainstream tools | Text description only |
| Specialized pose-transfer research tools | Yes, in some cases | Skeleton/keypoint mapping from a reference image — not typically a consumer product |
Exact reference-photo pose copying for stills exists mostly in specialized research tooling, not as a standard consumer AI photo generator feature.
For video, reference-based motion is a real, different capability — see Pose's native video generation with Motion Control.