Independent production field guide

Seedance 2.5 tutorial. Direct the shot.
Don’t describe the vibe.

A practical Seedance 2.5 tutorial for creators who want controllable motion, clearer camera direction, and prompts they can actually reuse.

Follow the tutorialBuild a prompt
  • Updated July 2026
  • Text-to-video + image-to-video
  • Free · no account
00:00–00:03Establish the place
00:03–00:07Give the subject one action
00:07–00:11Move the camera with purpose
00:11–ENDResolve the shot

Released July 2026

What's new in Seedance 2.5

Three capabilities that change how you write the prompt — not just how long the clip runs. Source: official Seedance 2.5 release ↗

Longer takes

30-second shots, multi-round extension

Single generations now run up to 30 seconds — double Seedance 2.0 — and you can extend a clip across multiple rounds while subject, scene, lighting, and sound stay consistent. What to change: write the extension as a continuation beat. Name what carries over, then the next action.

References

Up to 30 images, 10 videos, 10 audio

One generation can read appearance, motion, and sound from many references at once — ideal for multi-subject and multi-shot scenes. A white-model reference lets you block in camera, blocking, and light before any style is applied. What to change: stop redescribing appearance in prose and start directing how your references interact inside the frame.

Editing

Timestamp, green-screen, viewpoint edits

Edit a specific second of a clip, swap a background while keeping the subject and its physics, or re-direct the camera path — without regenerating from scratch. What to change: prompt the change and its time window, not the entire shot over again.

Here is what a multi-round extension prompt looks like — the structure 2.5's longer takes are designed for. Reference the previous clip, declare what must stay consistent, then direct only the next beat.

Extend @clip1. Continue the same subject, location, wardrobe, and visual style into the next 30 seconds; keep lighting and sound consistent. The boy with the football sprints down the train aisle; when the doors open he bolts onto the platform, the man chasing him through the crowd until he catches the boy's arm. Track beside them at a run, then settle the frame as the tension releases.

Three ways in

From first render to repeatable workflow

Start with the path that matches your current problem. Every page is written to stand on its own, then links to the next useful step.

Start here

Step-by-step tutorial

Turn an idea into a single directed shot, choose references, write the prompt, generate, and run a five-point QA pass.

Learn the workflow →
Steal the structure

Prompt examples

Browse practical prompts by purpose: product, cinematic, character, social, and image-to-video. Copy one, then change the controllable details.

Browse prompts →
Make it yours

Prompt template builder

Compose subject, action, camera, composition, environment, light, style, and sound without sending your text to a server.

Open the builder →

The useful mental model

A prompt is a tiny call sheet

Weak video prompts usually list attractive nouns: “a cinematic woman in Tokyo, neon lights, masterpiece.” That describes a poster, not a moving shot. A stronger prompt assigns a subject, a visible action, a camera decision, a setting that affects motion, and a clear ending. If sound matters, direct it as deliberately as the image.

A ceramic artist lifts a finished cup from the wheel and turns it toward the window. Slow handheld push-in from a medium shot to a close detail of the glaze. Quiet daylight studio, dust visible in the side light. Natural documentary texture. Audio: wheel slowing, soft room tone, one ceramic tap. End on the maker's hands holding the cup still.

The directing principles above stay useful across any video model. What Seedance 2.5 adds is what you can plan in a single prompt — longer takes, richer references, precise edits — so the sections below layer those capabilities onto the same call-sheet structure. Exact settings vary by product surface; confirm the options in your own interface.

REC00:00–00:12
The master prompt, storyboardedSlow push-in from medium shot to glaze detail — an original concept loop of the prompt on the left, not model output.

Production loop

Five decisions, in order

Fix the earliest weak decision instead of adding more adjectives at the end.

Choose one shot objective

Write what must be visibly different at the end of the clip. “Reveal the product label” is testable; “feel premium” is not.

Lock the subject and action

Give the subject one primary action. Multiple simultaneous actions increase ambiguity and make continuity harder to judge.

Choose camera behavior

Static, push-in, pull-back, orbit, pan, tracking, or handheld. Add a reason and endpoint instead of stacking moves.

Generate a clean baseline

Change one variable per iteration. A baseline makes it possible to learn whether the camera, action, or reference caused the improvement.

Review like an editor

Check identity, anatomy, object permanence, physics, camera path, audio sync, and whether the final frame resolves the promised shot.

Quick answers

Seedance 2.5 FAQ

What should a Seedance 2.5 prompt include?

At minimum: subject, visible action, camera behavior, composition, environment, lighting, and a clear endpoint. Add style and audio only when they change the shot.

Should I write one long paragraph?

A compact paragraph works for a single shot. For a longer sequence, use timed beats and keep one dominant action per beat. Avoid contradictory camera instructions.

Can I use these templates for image-to-video?

Yes. Treat the reference image as the source of appearance and use the prompt mainly to direct motion, camera, environmental effects, and the ending frame.

Is this the official Seedance website?

No. This is an independent field guide. It is not affiliated with or endorsed by ByteDance.

What can Seedance 2.5 do that 2.0 couldn't?

Three things, per the official release: single takes up to 30 seconds with multi-round extension; richer multimodal references (up to 30 images, 10 videos, 10 audio, plus white-model blocking); and precise editing by timestamp, green-screen, or viewpoint. The directing workflow stays the same — these expand what you can plan in one prompt.

Text to video and image to video

What this Seedance 2.5 tutorial helps you control

AI video generation is not one decision. It is a chain of decisions about what enters the frame, what changes, what stays consistent, how the viewer’s attention moves, and where the shot finishes. This guide turns that chain into a repeatable workflow. You can use it when starting from text, when animating a reference image, or when revising a shot that is almost right but fails in one important moment.

For text-to-video, begin with the visible event. Describe the subject and the action before you name a style. The model has to stage a scene that does not exist yet, so spatial relationships and a clear camera path matter. For image-to-video, the reference already defines much of the appearance. Your prompt should spend more attention on motion, timing, physical reactions, and which visual details must remain fixed.

Write for an editor, not a mood board

A production-ready prompt creates footage you can evaluate. It states what the shot should reveal, how the camera helps reveal it, and what must be visible in the final frame. If the output fails, you can trace that failure back to a specific instruction. That is the difference between adding random adjectives and directing a new attempt.

Keep an iteration log

Save the prompt, the output, and one sentence about the result. Record the single variable you change next: camera speed, subject action, reference choice, light, or endpoint. This small habit prevents circular prompting and gradually builds a private library of working structures. The examples on this site are starting points for that library, not magic phrases that guarantee identical outputs.

Next shot

Start with a prompt you can diagnose.

Or see how Seedance 2.5 compares to Sora 2 before you pick.

Build it now