Seedance 2.5 explained: sound, durations and what it costs

Seedance 2.5 is the model people mean when they say “AI video with sound”. It generates the audio track together with the picture — footsteps, ambience, a line of dialogue — instead of bolting sound on afterwards. Here is a practical guide: what it's good at, how it behaves, and how the pricing works.

What makes 2.5 different

• Native audio: sound is generated with the video, synced to motion
• Up to 1080p per request, 720p when you want it cheaper
• Scenes up to 30 seconds — long enough for a full beat, not just a loop
• Reference-driven: it follows a first frame closely, which is exactly what a frame-first workflow needs

How Actoria uses it

You don't write to an API. Compose the first frame, describe the action and camera, pick Seedance 2.5 in the model selector and set resolution and duration. Every scene is a separate generation, so you can mix models inside one project: 2.5 for the dialogue scene, the Wan 2.2 model for a cheap establishing shot. Finished scenes continue from their last frame like any other.

Sound: what to expect

Ambience and effects are consistently good — rain, traffic, a crowd, footsteps on gravel. Speech works best as short lines: write them in quotes in the prompt and keep the scene to one speaker. If a scene needs a soundscape the model didn't give you, Actoria can add a sound layer on top (sea, city, rain, or free text) — see AI video with sound.

Real faces

Seedance is strict about real people: a human face goes through only as a verified asset. In Actoria that is the liveness check — your own face, or a friend's via an invite link. Synthetic actors from the constructor or a prompt need nothing extra. This is a feature, not a limitation; it is what keeps real people in AI video consent-based.

Pricing, honestly

Seedance is billed per second of output, and resolution changes the price: 1080p costs nearly twice as much as 720p. Actoria's per-generation price is among the lowest of the platforms we surveyed — below every developer API we checked, though not the absolute cheapest consumer app. The full comparison across 15 platforms is in the Seedance pricing study, and the current per-second rates live on the Seedance page.

Prompting tips for 2.5

• One action per scene. “…then she turns and…” belongs in the next scene.
• Name the camera: slow push-in, handheld follow, static wide.
• Describe sound if it matters: “rain on a tin roof, distant thunder”.
• Quote dialogue, one speaker, under ten words.
• Keep wardrobe wording identical across scenes.