Seedance 2.5 Museum Clips: Parallax Without Reshaping
A collection-object video should not invent a different artifact while making the camera move. For Seedance 2.5, begin with a rights-cleared photograph and request a restrained viewpoint shift around a stationary object. Compare its contour and surface marks with the source. Generated motion is an interpretation of a photograph, not a conservation record or a verified reconstruction of hidden surfaces.
Key TakeawaysKeep the object still and request one small camera movement, rather than a complete orbit.Inspect rims, handles, decoration and surface marks throughout the clip; a convincing movement can conceal a changed object.Siray's Seedance 2.5 i2v schema accepts a source image and optional end image, but no dedicated camera-control field. Neither input guarantees faithful three-dimensional geometry.

Which Photograph Is Suitable?
Choose a view that shows the object's important features without glare or heavy obstruction. For a vessel, that may mean a readable rim, visible handle and clear relationship to its support. A dramatic photograph that hides half the object can be less useful than a straightforward documentary view for this exercise.
Separate permission to use a photograph from assumptions about the object. A freely viewable museum page does not automatically grant every reproduction right. Keep the source credit and any required usage conditions with the asset. If the identification is uncertain, do not turn a visual guess into an authoritative caption.
How Small Should the Viewpoint Change Be?
Start with a slight lateral movement and keep the object stationary. A full orbit asks the generator to supply surfaces absent from a single photograph. Even a small movement involves inference, but limiting the move narrows the visual change you need to inspect. It does not certify accuracy.
A plausible new angle may still change the handle attachment, painted motif or rim thickness. Judge those details against available reference material rather than accepting them because they look physically possible. Use the clip for an explicitly interpretive presentation, not as evidence of an unseen side.
An illustrative prompt:
A restrained lateral camera drift across the photographed ceramic vessel, with the object remaining stationary on its support. Keep the visible rim, handle attachment, decoration and surface marks consistent with the source. Subtle viewpoint change, stable lighting and quiet background. The vessel does not rotate, bend or change material.
Which Seedance 2.5 Fields Apply?
Use the general bytedance/seedance-2.5-i2v endpoint, not its Spicy sibling. As checked on October 15, 2026, the Siray Seedance 2.5 i2v specification requires model, prompt, image, duration, size and aspect_ratio.
Duration accepts integers from 4 through 30. Size choices are 720p, 480p and 1080p. The aspect-ratio enum contains adaptive, 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9. Select the intended framing deliberately and inspect the output; a supported value does not guarantee unchanged cropping.
end_image is an optional end-frame URL. It can specify a desired endpoint image, but it does not validate the intermediate geometry. audio_enable is optional with no declared default; request false explicitly for a silent visual study.
This schema has no seed, negative_prompt or structured camera-movement parameter. The Seedance camera-prompt guide covers general movement wording. Here, the narrower goal is inspecting an artifact's visible features during a restrained shift.
What Would Make the Clip Misleading?
Check the silhouette first, then the local features that identify the object. Look for changing handle thickness, drifting decoration, disappearing chips or a base that slides across the support. Review the full transition rather than only the best-looking final frame.
What if the object changes? Reject that segment. For exact preservation of the photographed view, a conventional pan or zoom over the original still may be more appropriate, although it does not create true parallax.
Can a second image prove the movement is accurate? Not by itself. Reference views can support comparison, but generated intermediate frames still need independent review.
Try a bounded interpretive study through Siray, keep the source photograph available, and label the resulting motion honestly.
Source: Siray, Seedance 2.5 i2v OpenAPI specification, retrieved October 15, 2026, linked above. No generation or geometric-accuracy test was performed.