If a subject is cut off in a Seedance 2.5 video, first identify where the loss occurs: the source image, the generated frames, the player preview, or a later export. Correct that stage before changing the prompt. A prompt cannot reliably restore a head, foot, or product edge that the source image never contained. Use a roomier reference, choose a matching output shape, and keep movement within the visible space.
Where does the cropping happen?
The symptom matters because each cause needs a different fix.
| Visible symptom | How to confirm | Priority action | Limitation |
|---|---|---|---|
| The original image already cuts off a head, foot, or product edge | Open the source file at full size before generation | Use a wider source image or extend its canvas in an external editor | Newly extended areas are generated content and must be checked |
| The source and target shapes differ | Compare the source ratio with the selected output ratio | Prepare a composition with more side or top/bottom room, then generate again | A prompt cannot change every missing pixel in the source |
| The opening is complete but motion pushes the subject outside frame | Compare opening, middle, and final frames | Reduce travel, push-in distance, or subject movement; generate another candidate | “Keep the full body” does not guarantee tracking |
| The download is complete but the preview is cropped | Open the downloaded file in another player or inspect its dimensions | Treat the player or export as a separate problem | A preview view is not proof that the file is cropped |
| A later reframing removes important details | Compare the edited version with the original export | Reframe with more padding or use the complete clip | Cropping cannot add information that was never recorded |
Do not treat every crop as a model failure. Source framing, aspect-ratio conversion, playback settings, and editing decisions can each create the same visual impression.
Which input photo is safest?
For a full-body person, include the entire head, feet, and a little space in the direction of the planned movement. Keep the camera perspective natural and avoid a severe wide-angle stretch at the edges. For a product, show the complete silhouette, including cap, base, handles, and any part that must remain identifiable. Leave room around the object instead of placing it against the image boundary.
When changing from landscape to vertical, do not simply ask the model to “fit everything.” Prepare a source where the important subject can survive the narrower composition. The image-to-video workflow can be used to start a new image-led test; the exact framing result still needs inspection.
Three practical ways to recover framing
A wider source image is usually the cleanest option. Keep the original subject smaller in frame, then generate with a modest action. This preserves real edges and reduces the amount of unseen space the system must invent.
If no wider source exists, expand the image canvas in an external image editor before uploading it. Treat the new border as a visual suggestion, not verified room geometry. Check hands, feet, labels, and straight architectural lines after generation.
For an existing clip, an external video editor can reframe, scale down, or add padding. This can protect a visible subject, but it cannot recover a foot or bottle edge that was already outside the exported pixels. Auto-reframe tools may help with a moving subject; Adobe documents this as an external editing workflow in its Auto Reframe overview, not as a Seedance control.
Example A: a full-body person with a small action
Reference preparation: Use a portrait or landscape photo showing the person from head to shoes, with clean separation from the background and extra space above and below. Do not use a tight crop and then request a wide camera move.
Teaching prompt example:
Use the supplied full-body reference image. Keep one adult person fully visible from head to shoes in a medium-wide composition with comfortable space above the head and below the feet. Preserve the person’s clothing, body proportions, and the background doorway. The person takes one small step in place and raises one hand briefly. Keep the camera nearly fixed and do not push in, orbit, or reframe. Keep the full body inside the frame from beginning to end.
Acceptance check: Pause at the opening, middle, and final frames. Confirm that the head, both feet, hand, and doorway remain visible. If the person leaves the frame, first reduce the step or remove the hand action; do not add several new camera instructions at once.
Example B: a tabletop product with movement space
Reference preparation: Use a product photo that shows the complete bottle, cap, base, label, and a clear section of tabletop around it. Keep the camera angle close to the intended video and avoid a source where the bottle touches the edge.
Teaching prompt example:
Use the supplied product photo as the visual reference. Keep the complete bottle centered on a matte tabletop, including the cap, base, label, and both side edges. Preserve the tabletop horizon and the neutral background. Allow a slow, very small lateral camera move while the bottle remains stationary. Keep enough empty space around the bottle for the move; do not zoom in, rotate around the product, or crop the cap or base. Maintain the same product proportions and label placement throughout.
Acceptance check: Inspect the cap, bottom edge, label corners, and the space into which the camera moves. If the cap disappears, simplify the camera move or start from a roomier source. If the label changes shape, treat that as a product-detail issue rather than a simple crop.
Is aspect ratio the same as resolution?
No. Aspect ratio describes the frame’s shape, such as landscape or vertical. Resolution describes the pixel dimensions inside that shape. A larger file can still have a poorly framed subject, and a correctly framed file can look soft for other reasons. For pixel dimensions, codecs, and blur diagnosis, see the Seedance 2.5 video-resolution guide. Keep this article’s framing checks separate from resolution troubleshooting.
How should you review a candidate?
First inspect the source image. Then review the generated file, not only the browser preview. Check the first, middle, and last clear frames, plus the moment when movement is largest. Compare the subject’s important boundaries with the source: head and feet for a person; cap, base, label, and handles for a product. Finally inspect any reframed export on the intended device.
FAQ
Why is my subject cut off in Seedance?
Check the source crop, target shape, and the point where the subject leaves the frame. Use a roomier source or reduce movement before changing other prompt details.
How do I keep a full body in an AI video?
Start with a source showing head to shoes and space around the person. Use a small action and restrained camera movement, then verify the opening, middle, and final frames.
How do I turn a landscape image into a vertical video without cutting off a product?
Prepare a wider or vertically composed source, or add external canvas before generation. A later editor can add padding, but it cannot restore a product edge absent from the file.
Can a prompt fix a cropped input image?
No. A prompt may describe the intended composition, but it cannot reliably reconstruct body or product areas missing from the input. Replace or expand the source, then inspect the generated result.
