Seedance 2.5 is now available in OpenVideoMaker. The new model extends the Seedance page with longer clips, native audio generation, and richer reference control. Open the Seedance generator and select Seedance 2.5 to move from a prompt, a reference image, a reference video, or a reference audio clip into a generated video with sound.
What is new
Seedance 2.5 brings four changes to the Seedance workflow. Clips now run from 4 to 30 seconds, so a single generation can cover a full product reveal or a longer social beat. The model generates audio in the same pass when you enable audio generation, so the output ships with a matching soundtrack instead of a silent track. Reference control accepts images, videos, and audio at the same time, letting you anchor subject, motion, and soundscape together. Resolution stays at 480p and 720p for fast, cost-conscious drafts.
How to prompt it
Treat the prompt like shot direction rather than a keyword list. Seedance 2.5 reads your intent first - it keeps who is in the scene, what happens, and how it ends intact, so your job is to organize the shot, not to pile on adjectives. Name the subject, the main action, the camera move, the lighting, and the mood in one focused scene; a single clear shot produces more coherent motion than stacking unrelated ideas.
A reliable prompt follows a clear structure: subject and scene first, then the main action, then the visual style, the camera move, and the sound. For example, "A girl in a red coat walks down a snowy street at dusk; the frame stays warm and cinematic; the camera tracks alongside her; soft wind and footsteps fill the audio." Drop any line you don't need, but the subject and the main action must stay clear.
When you upload references, give each asset one job and keep them from overlapping: images for the subject's look, clothing, or structure; videos for action, camera motion, or rhythm; audio for voice, ambient sound, or music. One asset, one purpose - don't ask the same image to define both a person and the lighting.
Seedance 2.5 runs up to 30 seconds, and duration is set on the page, so the prompt only organizes the event. For longer clips, write in stages where each stage holds one state change and ends on a concrete result, like "starts crouched, rises to standing, ends facing the window." Don't invent numeric timecodes the page never asked for; just extend the action, the reaction, or the transition.
Keep aspect ratio, total duration, resolution, frame rate, and the audio toggle out of the prompt - the page or the API sets those. The prompt is for the story, the look, and the motion, not the spec sheet. And don't over-constrain it: the model won't silently add quality packs, watermarks, logos, subtitles, or negative prompts you never asked for, so don't pad the prompt with them either. Say only what the scene truly needs.
Pricing preview
OpenVideoMaker estimates Seedance 2.5 cost in real time as you change duration, resolution, or upload media. The estimate is based on the output duration, resolution, and frame rate, with a different per-token rate when a reference video is attached. The final price is confirmed before you spend credits, and failed generations are refunded.
Try it
Start from the Seedance generator and pick Seedance 2.5, compare it with the broader AI Video Generator, or pair it with AI Image Generator when you need a clean still frame before animation.
FAQ
How do I use Seedance 2.5 as an AI video generator?
Open the Seedance 2.5 generator, describe one clear scene with the subject, action, camera movement, and mood, add image, video, or audio references when needed, choose the output settings, and generate a draft to refine.
What references can Seedance 2.5 use?
Seedance 2.5 can use image, video, and audio references. Give each asset one job, such as subject appearance, motion, camera rhythm, voice, or soundscape, so the prompt stays unambiguous.
Can Seedance 2.5 generate audio with the video?
Yes. Enable audio generation when the scene needs native sound, then describe the relevant voice, ambience, or music direction without turning the prompt into a separate technical specification.