03 · TUTORIAL · 9 MIN · SEEDANCE 2.0

The quickstart, from idea to first clip.

The shortest reliable path from a thought to a posted shot in Seedance 2.0. Seven numbered steps with copy-paste blocks — one set for your AI chat, one for the prompt box. The on-ramp for everything else here.

You needAn AI chat (ChatGPT, Claude, or DeepSeek)

HelpfulA way to make a still image — optional; many video tools include one

AndSeedance 2.0 access

Time~15 minutes for your first clip

Two colors run through this guide. Teal blocks get pasted into your AI chat — they make the AI do the thinking with you. Tungsten blocks get pasted into the Seedance video prompt field. You never have to invent wording from scratch. Fill the bracketed bits, copy, paste, go.

One rule for the teal blocks

Do all the teal (AI chat) steps in a single ongoing conversation — each one builds on the answer before it. The tungsten blocks go into Seedance.

Say your idea in one sentence

whyThe whole pipeline flows from one clear intention. A fuzzy idea makes a fuzzy video — so sharpen it before you touch any tool.

Write your idea as a single sentence. Then let the AI poke holes in it.

Paste to your AI chat

You are my cinematic video director. Here is my idea for one short
cinematic shot: [WRITE YOUR IDEA IN ONE SENTENCE].

Ask me three short questions that will most improve it as a single
continuous shot — about subject, setting, mood, or camera. Ask all
three at once, then wait for my answers. Don't write the prompt yet.

Turn it into a shot brief

whySeedance makes one continuous shot at a time. Lock the subject, setting, mood, light, and a single camera move now, and every later step gets easier.

Answer the AI's three questions, then send this:

Paste to your AI chat

Using my idea and answers above, write a tight shot brief for a single
continuous cinematic shot. Output exactly these lines and nothing else:

Subject:
Setting:
Mood:
Lighting (one line):
Camera (one move only):
Aspect ratio:

Keep each line short and concrete.

Make one clean source image

whyYour first frame decides the entire look. Seedance animates what you hand it — so a strong, clean still is most of the battle.

Get an image prompt from the AI, then run it in your image model. Three rules: vertical, subject big, nothing written on the image.

Paste to your AI chat

Write one image-generation prompt for the FIRST FRAME of this shot,
based on the brief above. Requirements: vertical 9:16 composition; the
subject fills 40–70% of the frame; cinematic lighting that matches the
brief. Absolutely no text, letters, numbers, arrows, captions, UI,
logos, or borders anywhere in the image. Describe only what is visible
in the frame. Output only the image prompt.

The #1 thing that breaks clips

Never feed Seedance an image with text, arrows, labels, or a collage of poses on it. The model can't tell notes from scenery — it renders your notes as floating objects. One clean frame, every time. (The deep version of this is the clean-source protocol.)

No image tool? You can skip this

Seedance also makes video from words alone (text-to-video). No image model handy? Jump to Step 4 and let the prompt build the scene. You trade some control for speed — a clean starting image still gives the steadier, more predictable result, so use one when you can.

Faces & people — important

Seedance 2.0 is built for scenes, places, products, and creatures, and it blocks uploads of realistic, identifiable real people. For a character, make one in an AI image tool (an invented face) and use that — never a photo of a real person. Need a specific real human? Use a model like Kling 3.0 or Veo instead.

Write the video prompt the Seedance way

whySeedance follows direction, not adjectives. There's an official shape that works: Subject → Action → Environment → Camera → Style → Constraints, kept to 60–100 words.

Let the AI build it for you in the right order:

Paste to your AI chat

Turn the shot brief into a Seedance 2.0 video prompt. Follow this exact
order: Subject, then Action, then Environment, then Camera, then Style,
then Constraints.

Rules: 60–100 words total; exactly ONE camera move; describe the camera
move and the subject's movement in SEPARATE sentences; include one
lighting line; no stacked adjectives (one strong word each); end with a
short "Avoid…" line listing jitter, bent limbs, and face warping.

Output only the prompt.

Here's a finished example so you can see the target. Paste it straight in, or use it as a model:

Paste into the video prompt field

A lone figure in a long coat stands at the edge of a rain-slick
rooftop. She slowly turns her head toward the skyline. Wide rooftop at
night, distant neon haze, light rain falling. The camera pushes in
slowly, then holds. Cool cinematic light, rim light from the signs
behind her. Filmic, sharp focus, vertical 9:16.
Avoid jitter, bent limbs, and face warping.

Generate

whySettings that match the model's defaults give the steadiest result. Don't fight the engine on your first clip.

In Seedance: pick your aspect ratio, upload the clean image, paste the tungsten prompt, and generate. A few things worth knowing so nothing surprises you:

Length

About 4–15 seconds depending on surface — some set it automatically from how much you describe, others let you pick (5/10/15s). One simple action ≈ short. Want longer? That's the advanced guide.

Look

720p by default, 1080p ceiling. A clip takes ~1–3 minutes to render. Audio is generated built-in — ambient sound and music come for free.

Prompt size

Keep it tight — 60–100 words, a few sentences. Past ~100 words, quality drops as the model's attention spreads thin. Short and clear wins.

Judge it, then fix one thing

whyChanging one variable at a time is the only way to learn what actually moved the result. Resist the urge to rewrite everything.

Watch it once and ask two questions. Is the subject steady and undistorted? Is the motion one smooth move, not shaky? If something's off, match it below:

Shaky / jittery

You mixed camera and subject motion, or used the word "fast." Put them in separate sentences; cut "fast."

Subject melts or warps

Your source frame was busy, or you re-described the subject too much. Start from a cleaner image; let it carry the look.

Floating text / weird shapes

Your source image had text or marks on it. Remake it fully clean — the prompt can't remove them.

Flat / not cinematic

Add one lighting line. Lighting is the single highest-leverage thing you can add.

Then make exactly one change with the AI:

Paste to your AI chat

Here is my Seedance prompt: [PASTE PROMPT]

The problem with the result was: [DESCRIBE THE ONE ISSUE].

Change only the single element that fixes this and keep everything else
identical. Say what you changed in one line, then output the revised
prompt.

Publish

whyShipping beats polishing. A posted clip teaches you more than a tenth re-render ever will.

Export at your platform's aspect ratio and post it. That's a real cinematic shot, start to finish. When one shot isn't enough — a recurring character, a story, something longer — the next guide is waiting.

Keep this open

The whole method, on one screen

The rules that matter

✓One clean image, no text or marks on it, ever.
✓One camera move per clip. Describe camera and subject in separate sentences.
✓Add one lighting line — it's the biggest quality lever you have.
✓Never use "fast" + busy scene + quick cuts together. Pick one.
✓Keep prompts 60–100 words. One strong adjective, not three.
✓End character prompts with "Avoid jitter and bent limbs."
✓When fixing, change one thing at a time.
✓Seedance is for scenes, products & invented characters — not real faces.
✓Do the teal (AI) steps in one ongoing chat.

Seedance 2.0 at a glance

The numbers, for when you need them

4–15s

per clip · varies by surface

reference images max

video refs (≤15s total)

audio refs (≤15s total)

files total, across all types

60–100

words is the prompt sweet spot

Built for scenes, products, and AI-made characters — realistic real-person faces are restricted. Audio is generated built-in. Go longer with Extend or by chaining clips (advanced). Limits and resolution (720p default, 1080p ceiling) vary by where you run Seedance — CapCut/Pippit, Dreamina, fal.ai differ. Verified mid-2026; check your platform's current page.

When one shot isn't enough

Direct the Scene — the advanced guide