Upload the document
Tell the model: “Use this as the Seedance 2.0 prompting knowledge base.”
Prompt assistant setup
Build your Seedance 2.0 prompt engine.
Turn the master guide into a reusable Custom GPT or upload it as knowledge to any capable LLM. The goal is simple: give the AI a clear directing framework so it can write stronger Seedance prompts on demand.
Knowledge beats memory
A good assistant does not improvise from vague taste. It follows a reference system, asks for missing inputs, and writes shot-ready prompts.
How to create the Custom GPT
Use the document as the knowledge base, then instruct the GPT to behave like a Seedance video director and prompt engineer.
01
Create a new GPT
Open the GPT builder, create a new assistant, and name it something clear like Seedance 2.0 Prompt Director.
02
Upload the knowledge document
Add the Custom GPT document as knowledge. This gives the assistant the prompt formula, shot-script structure, reference rules, camera movements, constraints, and troubleshooting logic.
03
Add clear instructions
Tell the GPT to ask for the video format, references, duration, aspect ratio, platform, character details, style, and goal before writing the final prompt.
04
Force structured output
Ask it to return prompts in Seedance-ready format: style, duration, timecoded shots, reference roles, consistency rules, and negative constraints.
05
Test and refine
Run a few examples: fight scene, product UGC, transformation, POV, and animation. If the prompts are too broad, instruct it to simplify action and camera hierarchy.
System instructions you can paste
A concise instruction block for your Custom GPT or any AI assistant.
You are a Seedance 2.0 prompt director. Your job is to write production-quality AI video prompts using structured scene logic, clear reference roles, physical action, camera hierarchy, and negative constraints. Before writing the final prompt, ask for missing inputs: format, references, duration, aspect ratio, character identity, environment, style, action, audio needs, and output platform. Prefer shot-script prompts for complex scenes. Keep every shot readable, physical, and cinematic. Avoid prompt overload, unassigned references, too many camera moves, identity drift, temporal flicker, bent limbs, and chaotic composition.
Use it with any LLM
You do not need a Custom GPT. You can use the document with ChatGPT, Claude, Gemini, or another LLM if it supports file upload or long context.
Describe the idea
Give it your format, characters, references, style, platform, aspect ratio, and duration.
Ask for a shot script
Request a Seedance-ready prompt with timecodes, reference roles, consistency rules, and avoid list.
The quality checklist
Use this before you paste any prompt into Seedance 2.0.
Every reference has one job
Character, environment, product, motion, camera, style, audio, or rhythm. Do not leave references ambiguous.
Every shot has one main action
If the shot tries to do everything, the generation usually becomes unstable.
Camera is simple
Use one primary move, then let the subject action carry the scene.
Constraints are explicit
Include avoid jitter, bent limbs, identity drift, temporal flicker, warped hands, and chaotic composition when relevant.