Mixer AI
How to use Seedance 2.5 for footage that looks filmed
Realistic video from Seedance 2.5 comes from describing the scene like a shoot: who stands where, what happens second by second, where the light comes from and what we hear. Pick the mode that fits your source files, give each file a role, and test the idea as a 480p draft before finishing it in 1080p.
Step by step
Pick a mode.
From text needs no files. From frames uses one image as the first frame, or two as the first and last frames. From references takes up to 30 images, 10 videos and 10 audio files. Video editor takes one finished video of 4–30 seconds.
Upload references and name each one in the prompt: “image 1 is the woman's face, image 2 is the kitchen, video 1 is the camera move”.
Audio is accepted only together with an image or video.
Write the scene in order: file roles, shooting style, place and light, timecoded actions, dialogue, camera, sound and how the scene ends.
Finish with a line listing what must not appear.
Set a duration from 4 to 30 seconds and the aspect ratio, then turn on Draft: the clip comes back in 480p and costs less.
The price is shown before you start.
Review the draft: camera, action, voice and sound.
If it works, click Complete draft within 7 days to get a 1080p version with the same prompt and files. Small details may shift slightly during completion.
Ready-to-use Seedance 2.5 prompts
The first prompt is written for From references, the second for From text and the third for Video editor. Replace the file numbers with your own.
Image 1 is the woman's face and clothing. Image 2 is the kitchen. Handheld documentary camera. Evening, warm light from a lamp above the table, dark outside the window. 0–3 s: medium shot, the woman stands at the stove with her back to the camera, stirring soup. 3–7 s: close-up, she turns around and gives a tired smile. She says quietly and slowly: “Dinner in ten minutes.” 7–10 s: medium shot, she turns off the stove and leans on the table. Visible skin pores, loose strands of hair, faint shadows under the eyes. Sound: soup bubbling, a quiet fridge hum, her voice. No music, no other people, no plastic-looking skin.
One continuous shot, a handheld camera follows a man in his forties wearing a grey coat. Wet pavement after rain, shop windows and street lamps reflected in the puddles. The man walks at a steady pace, turns up his collar and puts his hands in his pockets. The camera sways slightly and lags a little behind him. Natural grain, muted colours. Sound: footsteps on wet asphalt, a passing car, distant city noise. No music, no on-screen text, few passers-by.
Video 1 is the source clip to edit. Image 1 is the new jacket. Keep the framing, camera movement, the person's motion and facial expressions, the timing of every action and the sound of video 1. Replace only the person's jacket with the jacket from image 1: same cut, colour and fabric texture. Face, hair, background and lighting stay unchanged. No new objects or people.
Price in coins
Seedance 2.5 video: which mode to choose
| Task | Mode | What to upload | What to write in the prompt |
|---|---|---|---|
| A scene from scratch | From text | Nothing | Place, people, actions by the second, light and sound |
| Animate a frame or reach a set ending | From frames | 1 image as the first frame; 2 images as first and last frames | What changes from the first frame to the last |
| Your own characters, location, camera move, voice | From references | Up to 30 images, 10 videos, 10 audio files; audio only with an image or video | The role of every file: image 1 is the face, video 1 is the camera, audio 1 is the voice |
| Change a finished clip | Video editor | One 4–30 second video, plus optional images and audio | What stays as it is and what changes |
In Video editor the result takes its length and aspect ratio from the source video. Seedance 2.5 has no extension and no dedicated motion-transfer mode.
Realistic Seedance video: what to put in the prompt
Open with the shooting style: handheld camera, locked-off tripod, security camera in a top corner, 35 mm film. That line shapes every other choice.
Describe skin and small flaws: pores, fine lines, uneven tone, flyaway hair, light grain. Perfect symmetry and a glossy finish give the AI away.
Light should come from a source inside the scene: a window, a street lamp, a low sun. A studio-lit photo of your character looks pasted onto a street background, so prepare it in the scene's light.
Keep faces large with close-ups or medium shots. One or two characters, one or two actions, a calm camera. End with an exclusions line: no plastic skin, no blown highlights, no oversaturated colours, no music.
Timecoded shot changes
Write several shots of one scene as consecutive segments: “0–2 s: close-up of a ticking clock. 2–6 s: the man at his desk, side view.” Segments have no gaps or overlaps and add up to the clip length.
Choose each segment's length by framing and action: a detail is usually shorter, a wide shot with action runs longer. Use one camera move per segment.
An object or place described only in text stays the same only inside one generation. If each shot needs its own retake, make shots as separate generations with the same references.
Reference roles in the prompt
The model does not guess a file's purpose from upload order. Name every file: “image 1 is Mira's face, image 2 is the bar, video 1 is the camera move, audio 1 is her voice”. A file without a role may be ignored.
Prepare a character sheet for each person: one person in three or four views on a white background. Leave small accessories off the sheet unless they must appear: in motion they blur or duplicate.
If there must be only one of something, say so: “exactly one cup, never a second one”. You do not need to fill all thirty image slots: start with the few files that matter.
Dialogue and sound
Sound is on by default in Seedance 2.5. Write each line as speaker, tone and pace, then the exact words in quotes. The spoken language follows the text of the line.
Lip movement can drift on non-English speech, Russian in particular. Keep lines short, leave them extra time and write numbers as words.
Name sounds together with their actions: a glass hits the table, steps on gravel. Describe ambience and music separately, and write “no music” when you want none. A wrong voice or an extra line is audible in the draft, so check it before completing.
What Seedance 2.5 handles less well
Small on-screen text, phone and laptop screens with fine detail, exact words in motion graphics. Check a short caption letter by letter; add longer text over the finished clip.
Keeping the first frame exactly: the model may change the angle or pose or add a small action. State positions and framing explicitly and regenerate the failed piece.
Small faces in wide shots, finger counts, a profile without a side reference, crowds of separate characters. Reduce the number of people per shot and split a long take with large fast movement into shorter generations.
Which model to use
FAQ
How do I use Seedance 2.5?
Pick the mode that fits your files, name each file's role in the prompt, describe actions by the second and the sound, and turn on Draft. Finish a good draft in 1080p with Complete draft.
How long can a Seedance 2.5 video be?
From 4 to 30 seconds per generation. In Video editor the result is as long as the source clip.
Can I write Seedance 2.5 prompts in languages other than English?
Yes. If the clip has speech, keep lines short and write numbers as words so lip movement drifts less.
Can Seedance 2.5 extend a video?
No, Seedance 2.5 has no extension. Make the next part as a new generation: use the last frame as the first frame in From frames, or the previous clip as a video reference.
What is the difference between a draft and the final video?
A draft is made in 480p and costs less. Completion to 1080p keeps the same prompt and files and can be started within 7 days.