First Frame vs Reference Image in Seedance 2.5

Choose what the picture must do

A first frame supplies a starting composition. A reference supplies visual guidance without requiring that opening frame. Neither guarantees unchanged identity, lettering, geometry or camera movement. Decide which requirement matters before adjusting the prompt.

Match image is specific to first-frame mode

Seedance 2.5 image mode uses the source image’s ratio through the Match image setting. Reference mode offers selectable ratios and Automatic aspect ratio. Both automatic settings use the parameter value adaptive, but they serve different input roles; do not read Automatic aspect ratio as a promise to preserve the reference image’s dimensions.

Keep the requested value unchanged when comparing modes, then record each video’s actual dimensions. For vertical delivery, prepare a vertical first frame or select a supported reference-mode ratio. Neither guarantees an exact crop or subject position.

Use one source image for both requests

The source is our amber perfume bottle on a black stone plinth: a 1536 × 1024 WebP, generated for this site on 19 October 2026 without a reference input. We uploaded it once and reused the same processed image asset for both requests.

Keep the original beside both videos. Check the cap, glass outline, amber material, bottle position and plinth edge. A convincing new composition may still miss the required opening layout.

Keep the prompt and exposed settings the same

For both requests we selected Seedance 2.5, four seconds, 480p, adaptive aspect ratio, native audio off and draft off. Only the input mode changed. The prompt asks for a stationary bottle and sideways camera movement.

One continuous four-second shot. The amber perfume bottle stays on the black stone plinth. The camera slowly tracks sideways to reveal the glass edge while the bottle remains fully visible. A thin wisp of smoke drifts in the background. Preserve the bottle shape, cap, amber glass, warm side lighting and dark setting. No cuts, no new objects, no text.

What the two actual outputs show

The first-frame output keeps the source’s approximately 3:2 layout. The whole bottle and cap are visible with space above the cap in the reviewed frames. The right side of the glass and plinth becomes more visible as the smoke changes; this is consistent with a slight viewpoint shift, without proving an exact camera path.

The reference output is a wider 16:9 composition. Its glass bottle body remains visible, but the top of the cap is outside the frame at the beginning, middle and end. It therefore misses the prompt’s request to keep the entire bottle visible. The amber glass, stone plinth and warm light remain recognizable. We kept this imperfect result rather than generating a replacement.

The same 480p preset returned different pixel dimensions. For a shot that needs this specific opening layout, the first-frame result is the closer match in this pair. A new request may differ; neither input mode is a guarantee of identical framing or geometry.

Open a setup, then attach the image

These links prefill the settings and the same prompt. They do not upload an image or submit a generation. Attach the source, confirm its preview and mode, then review the current quote.

Uploads support JPG, PNG or WebP up to 10 MB, with each dimension from 300 to 6000 pixels and width divided by height from 0.4 to 2.5. This workflow has one image slot, with no separate last-frame upload.

Each completed example used 70 studio credits, 140 in total. These are site credits, not a cash price or a measured model-provider bill. Generation requires an eligible verified account and sufficient balance. Purchases are currently closed; you can prepare the prompt and inspect settings, but a new account is not a promise of free generation.

Judge the whole clip against the brief

Compare the opening with the original, then watch the middle and end for changes in the cap, silhouette and material. Check whether the entire bottle stays visible. Separate camera movement from object movement: a changing viewpoint is different from a bottle sliding, turning or changing shape.

Record the downloaded files’ actual duration and dimensions. Note any cuts, new objects or text and when they appear. Keep both outputs, including an imperfect one. One attractive frame does not show whether the full video works.

Make the next decision before another attempt

If the opening composition matters and reference mode rearranges it, reconsider the input role before rewriting the prompt. If movement is missing, decide whether you wanted a moving subject or a moving camera. Remove competing pan, orbit and zoom directions; a fixed-camera instruction remains a request, not a guaranteed lock.

Write one observed issue and one proposed change before spending more credits. Stop when the budget is exhausted or the job is still running or needs review. Inspect that existing job instead of submitting a duplicate.

Questions and answers