Seedance 1.5 Pro vs Seedance 2.5: The Cheap One and the Long One

Seedance 2.5 is the default video model on Treza. Seedance 1.5 Pro is the cheapest model in the Seedance family we carry. The version numbers imply a ladder, and the ladder is misleading: 1.5 Pro does several things 2.5 cannot do at all, including 1080p output and reference images. Choosing between them by version number will either cost you money on drafts or cost you a shot you needed.

This is a comparison of what each model actually accepts and what each one actually charges. Every spec below is read off Treza's video model catalog as of September 2026, and every cost figure is measured from real runs rather than estimated from marketing copy.

The short version

• Seedance 1.5 Pro is the drafting and detail model. It goes up to 1080p, offers seven aspect ratios including 9:21, accepts reference images, and is by a wide margin the cheapest way to look at a prompt. Its ceiling is 12 seconds per clip.
• Seedance 2.5 is the production model for long single takes. Any exact duration from 4 to 30 seconds, native audio at no extra token cost, six aspect ratios. It tops out at 720p and does not take reference images.

Most pipelines that use one end up using both.

Parameter comparison

Two rows in that table are the whole decision.

Length is the reason 2.5 is the default

Seedance 2.5 renders any exact duration from 4 to 30 seconds in one generation. That is not a convenience. A faceless channel pipeline writes a narration line, sends it to text to speech, and gets back an audio clip of some specific, awkward length. If your video model only does 5 and 10 second steps, you cut the writing to fit the model. If it does exact seconds up to 30, the shot matches the line and nothing gets trimmed to hide a seam.

That is why 2.5 is what our own AI video pipelines render on, and it is the reason a Short can come out of one video node instead of three plus a stitch. A Sequence node will stitch up to 12 shots, so 30 second takes are not a hard requirement, but every stitch you avoid is a cut you do not have to hide.

Seedance 1.5 Pro caps at 12 seconds. For a talking-head-free b-roll beat that is plenty. For a full narrated segment it is not, and no amount of price advantage buys back a shot the model cannot render.

Price is the reason 1.5 Pro exists

Both models bill in ByteDance's opaque video_tokens, so the published SKU rate does not translate into a price you can quote. Token counts scale with pixels and seconds, which means the only honest way to price these is to run them and read the invoice. We did that, because our credit gate has to reserve a sane amount before a run starts.

Measured across real 4 and 5 second runs at 9:16 with audio off, Seedance 1.5 Pro came in at about $0.026 per second of provider cost. A 5 second clip in that configuration is roughly thirteen cents of provider cost before our markup. Seedance 2.5 measured at roughly ten times that per second.

Our estimator does not use the measured floor. It reserves at $0.12 per second for 1.5 Pro and $0.25 per second for 2.5, because the reserve has to cover the worst configuration each model can be asked for: audio on, 1080p where the model offers it. Runs settle afterward at actual provider cost plus markup, so the reserve is a hold, not a price. For a like-for-like ceiling comparison, 2.5 costs about twice 1.5 Pro per second. For the configuration most people actually draft in, the gap is much wider than that.

One detail worth knowing before you turn audio on. Seedance 1.5 Pro publishes two SKUs, one with audio and one without, and the audio rate is exactly double. Seedance 2.5 publishes the same rate either way. So on 1.5 Pro, generating silent drafts genuinely halves the bill, and on 2.5 there is no reason to turn audio off. If you have been leaving audio on out of habit while iterating on 1.5 Pro, that is a fifty percent saving sitting in a checkbox.

For the wider picture on what a whole video costs once narration, transcription, and captions are in the graph, we published the full cost breakdown separately.

Reference images: the thing 1.5 Pro has and 2.5 does not

OpenRouter's video API exposes two different ways to feed images into a generation, and they are frequently confused.

frame_images pins an exact first or last frame. Both Seedance models support both ends, so both can open on your logo art and land on your product shot.

input_references is different. It passes several images as subject, style, or identity guidance rather than as literal frames, which is what keeps a character or a product recognizable across separate generations. Seedance 1.5 Pro accepts them. Seedance 2.5 does not, and falls back to keyframes.

That routing is confirmed by probe, not by reading the provider's description, which for 1.5 Pro never mentions references at all. We submit one reachable reference image and one unreachable one: a model that reads past the first fails, one that honors only a single reference completes. Failed jobs report no usage, so the probe costs nothing. Seedance 1.5 Pro read past the first.

The practical consequence: if you need the same character in shot 3 and shot 9, that is 1.5 Pro or Seedance 2.0, not 2.5.

Resolution goes the wrong way

Seedance 1.5 Pro outputs up to 1080p. Seedance 2.5 stops at 720p. The newer, longer, more expensive model has the lower ceiling.

For phone-first vertical content this rarely matters. It matters when the asset gets shown large: a site hero, a paid placement, anything on a desktop screen. In that case the usual move is to draft on the cheap model, pick the winning prompt, and re-render on something with headroom. Veo 3.1 goes to 4K. On the canvas that switch is one dropdown on the video node, and the rest of the graph does not change.

Seedance 1.5 Pro also carries 9:21, the extra-tall vertical ratio, which 2.5 does not. It is a narrow use case, but if you need it, only one of these two models has it.

How to actually use both

The pattern that works, and the one our own pipelines follow:

1. Draft on Seedance 1.5 Pro, 720p, audio off. At roughly three cents of provider cost per second, running five prompt variants is not a budget decision.
2. Look at the outputs. Kill the framings that do not work. This is where the money is saved, because the expensive model never sees a prompt you have not already looked at.
3. Render the keeper on Seedance 2.5 at the exact length your narration needs, audio on, since it costs nothing extra there.
4. If the destination shows the asset large, render the keeper on 1.5 Pro at 1080p instead, or move up to Veo 3.1.

You can also stop at step one. If your shots are all under 12 seconds and a text to speech node is already providing narration, 1.5 Pro is the whole pipeline at a fraction of the cost. That is especially true for short-form output, where the 9:16 crop and the 720p ceiling remove most of what 1.5 Pro gives up and leave only the price difference. The AI video generator exposes the same model dropdown as the canvas, so testing it takes a minute rather than a migration.

What we are not claiming

We have not benchmarked prompt adherence, motion quality, or physical plausibility between these two models, so this post does not rank them on any of that. Everything above is either a parameter the catalog publishes or a cost measured from our own invoices.

Frequently Asked Questions

Is Seedance 2.5 better than Seedance 1.5 Pro?

Not in a way the version numbers explain. Seedance 2.5 renders far longer clips, up to 30 seconds against 12, and it is our default for that reason. Seedance 1.5 Pro outputs higher resolution, up to 1080p against 720p, accepts reference images, offers one more aspect ratio, and costs substantially less per second. They are different tools rather than two rungs of a ladder.

Which Seedance model is cheapest?

Seedance 1.5 Pro, by a wide margin. Measured across real 4 and 5 second runs at 9:16 with audio off, it came in around $0.026 per second of provider cost, roughly a tenth of what Seedance 2.5 measured in the same exercise. Our credit gate reserves more than that, at $0.12 per second, because the reserve has to cover 1080p with audio on. Runs settle at actual cost afterward.

Does turning off audio save money on Seedance?

On Seedance 1.5 Pro, yes. The catalog publishes a with-audio rate exactly double the without-audio rate, so silent drafts cost half. On Seedance 2.5 the published rate is identical either way, so leaving audio on costs nothing.

Can Seedance 2.5 use reference images?

No. It supports first and last frame keyframes, but not the multi-image reference input that keeps a subject consistent across separate generations. Seedance 1.5 Pro and Seedance 2.0 both accept references, so use one of those when a character or product has to stay recognizable shot to shot.

What resolution does Seedance 1.5 Pro output?

480p, 720p, or 1080p. That is a higher ceiling than Seedance 2.5, which stops at 720p. If you need more than 1080p, Veo 3.1 goes to 4K, and switching is one dropdown on the video node.

How long can each model generate?

Seedance 1.5 Pro renders any exact duration from 4 to 12 seconds. Seedance 2.5 renders any exact duration from 4 to 30 seconds. Both let you pick the exact second rather than a fixed step, which is what lets a clip match a narration track instead of the other way around.