Short answer: Use Seedance 2.5 first last frame mode when the opening and closing images of the shot must be exact, such as a product reveal that has to end on a packshot. Use reference mode when the identity, product or style must stay consistent but the camera and composition can change. Pick one per generation: in the Seedance 2.5 API documentation we checked, first/last frame images cannot be combined with reference images or reference videos in the same request.
This guide is for creators who already have the images and are deciding which Seedance 2.5 input mode to use. It compares what each mode locks, what it leaves to the model, and the mistakes that most often waste a generation.
Seedance 2.5 first last frame vs reference mode at a glance
| Mode | What you upload | What it locks | Best for | Common failure |
|---|---|---|---|---|
| First frame (image-to-video) | One image | The opening frame | Animating a finished hero image or product shot | The ending drifts because nothing fixes it |
| First and last frame | Two images: a first and a last | The opening and the closing frame | Reveals, transformations, before/after shots, a fixed end card | Two frames that do not match in size, framing or lighting, so the motion between them looks forced |
| Reference mode | Images, videos and/or audio as references | Who or what appears, and the look; not exact frame positions | Keeping a character, product or brand style consistent across new compositions | Expecting a reference image to appear pixel-exact as the first or last frame |
Limits for reference mode on our Seedance 2.5 page: up to 50 references in one generation (30 images, 10 videos and 10 audio files), 4–30 second output, and 480p, 720p or 1080p. Check the current settings on the Seedance 2.5 AI video generator before you plan a shot.
How Seedance 2.5 first last frame mode works
In first last frame mode you give the model a fixed start and a fixed end, and it creates the motion between them. Runware's Seedance 2.5 API documentation describes the rules this way (paraphrased):
- Frame images take one or two entries. One image becomes the first frame; two images become the first and the last frame.
- A first frame is required. A last frame on its own is rejected, because the model needs a starting point.
- Two frames is the cap. For more beats in one shot, the documentation points to reference images used as keyframes instead.
- Give the first and last frame the same dimensions; otherwise the last frame is stretched to fit.
- Frame images cannot be used together with reference images, reference videos or an input video in the same request.
Because the frames are fixed, the output takes its geometry from your images. Build both frames at the aspect ratio you want to publish, with the same camera height, lens feel and lighting, and describe only the change that happens between them in the prompt.
For preparing a strong opening image, see our Seedance 2.5 image-to-video guide.
How Seedance 2.5 reference mode works
Reference mode does not pin a frame. It tells the model who or what must appear and how it should look, and leaves the composition to the prompt. ByteDance Seed describes Seedance 2.5 as "built for 30-second storytelling with precise reference control and powerful editing capabilities." In reference mode you can mix image, video and audio references and address them in the prompt by order, for example @Image1 for the product and @Video1 for the camera move.
Runware's documentation also notes that several stills passed as references can be prompted as keyframes the model moves through in order, but that such keyframes "guide rather than pin": the model approximates each still instead of matching it exactly. That is the main trade-off against first last frame mode.
For a full walkthrough of reference roles, read our reference to video AI guide.
Which Seedance 2.5 mode to choose by job
The last frame must be exact: use first last frame
A product ending on its packshot, a logo card, a sketch turning into a photo, or a day-to-night change of the same scene. If you would reject the clip when the final frame differs from your image, use first and last frame.
The subject must stay consistent, the shot can change: use reference mode
A recurring character in a new scene, a product shown from new angles, or a brand look applied to a new idea. Here a pinned frame would only limit the camera.
You have more than two story beats: use reference images as keyframes
First last frame mode is capped at two frames. For three or more beats, pass the stills as references and state their order in the prompt, accepting that each beat is approximated rather than locked.
You have one good image and no fixed ending: use first frame only
Upload one image as the start frame and describe one clear motion. Add a last frame later only if the ending keeps drifting.
How to test Seedance 2.5 first last frame vs reference mode
- Write down what must not change. If it is a frame, use frame mode. If it is a subject or style, use reference mode.
- Prepare matching frames. For first and last frame, export both images at the same size and aspect ratio.
- Start short. Run a short, low-resolution test first and check the credit estimate in the generator before each run.
- Change one variable per run. Adjust the prompt or swap one image, not both.
- Do not mix modes. If you need both a fixed ending and a consistent character, put the character into the two frame images themselves.
You can run these tests in the browser on the Seedance 2.5 generator; choose the input mode it offers and review the estimated credits before you generate.
Seedance 2.5 first last frame FAQ
Can I use Seedance 2.5 first last frame and reference images together?
Not in the same request, according to the Seedance 2.5 API documentation we checked. Frame images cannot be combined with reference images, reference videos or an input video. Choose one mode per generation.
Can I give Seedance 2.5 only a last frame?
No. The API documentation we checked says a first frame is required and a last frame on its own is rejected. Provide a first frame, or a first and a last frame.
How many frames does Seedance 2.5 first last frame mode accept?
Two at most: one first frame and one last frame. For more beats, pass the images as references and describe their order in the prompt.
Is Seedance 2.5 reference mode better for character consistency?
Usually, when the character must appear in new compositions. Reference mode keeps who appears and how they look while letting the camera and scene change. If the exact opening and ending images matter more, use first last frame mode.
How long can a Seedance 2.5 video be in either mode?
ByteDance Seed says Seedance 2.5 can "Create videos up to 30 seconds in a single generation, with the option to extend twice". Our Seedance 2.5 page lists 4–30 second output. Available settings can differ by input mode, so check the generator before you plan the shot.
Sources
Checked October 14, 2026 and named without links: ByteDance Seed's official Seedance 2.5 page (quotes above); Runware's Seedance 2.5 API reference and its "Image to video and keyframes" guide (frame-image rules, paraphrased); the Seeddance API "Models and inputs" documentation (one image as start frame, two as first and last frame, three or more as reference); our own Seedance 2.5 page (durations, resolutions, reference limits). Provider API rules can differ from the in-browser generator; the generator shows the modes available to you.
Try Seedance 2.5 Today
Ready to experience the next generation of AI video creation? Try Seedance 2.5 for free and see the difference for yourself.
