Seedance in Practice: AI Video Techniques

Seedance 2.0 gives you full control over the character, their clothing, the location and the action. You upload separate references - a character map, a clothing collage and a background - and the model combines them into a single cinematic video, holding the character's appearance steady in every frame.

In this lecture you'll build a character map in Nano Banana from your own photos, prepare a clothing collage and a location reference, and turn a short idea into a full Seedance prompt with the help of ChatGPT or Claude. Then you'll set up the generation, upload the references in the right order and link the tags so the model knows which reference belongs to what.

A useful shortcut for the clothing collage: instead of spelling out every item, you can name a group - "a full outfit and accessories for a Formula racing driver", for example - and the model assembles a coherent set on its own.

All prompts used in this lecture are in the attached resource file.

Elements is a Seedance feature that lets you assemble a cinematic scene from separate pieces: a character, locations and other objects. You create a reference for each element individually, and the model combines them into a single video, holding each object's appearance steady in every frame.

In this lecture you'll identify the key elements of your scene, write map prompts for each of them with Claude or ChatGPT and generate the maps in Nano Banana. Then you'll write the video prompt from a ready-made template, set up the generation and link the tags correctly.

You'll also learn why every action should be described as concretely as possible ("the character turns his head left, then right" instead of "the character surveys the area"), why copying and pasting the prompt breaks linked tags, and how to get around a blocked generation with a grid or a pencil veil on the reference.

All prompts used in this lecture are in the attached resource file.

Seedance can produce complete videos with a considered plot using a storyboard - a logical sequence of the key frames of the whole story. You control both the story itself and how it looks, including the characters' appearance.

In this lecture you'll get a character map prompt from Claude, generate the map in Nano Banana, and then build a full storyboard in GPT Image from a short idea and the map. Finally, you'll bring the storyboard to life in Seedance with a short prompt - the frames already carry the whole dramaturgy, so nothing else needs describing.

If a particular frame of the storyboard isn't right, you don't have to redo everything: you'll see how to ask Claude for targeted edits to just that part of the prompt.

All prompts used in this lecture, including the full storyboard instruction, are in the attached resource file.

Seedance 2.0 turns a child's drawing into an animated scene and a full video, preserving the distinctive features of the original artwork. Two paths are covered: a static frame with Nano Banana, and animation generated directly from the drawing in Seedance.

In this lecture you'll first turn the drawing into a 3D cartoon frame to see how the model reads it, then set up video generation in Seedance with the drawing as the anchor for the characters' appearance and the composition of the scene.

You'll learn the simple four-part animation prompt: the style of the scene, the elements that enrich the environment, the mandatory requirement to preserve the characters' appearance, and a short description of the action and camera movement. And you'll see how to handle unsuccessful generations without changing everything at once.

The example prompts from this lecture are in the attached resource file.

Video extension in Seedance 2.0 adds new footage to an existing video, smoothly continuing the original scene. You cut out the last fragment of the video, describe what happens next, and the model generates the continuation while preserving the characters and the visual style.

In this lecture you'll learn how to cut the right fragment (at least 4 seconds, with every important element clearly visible), prepare optional references such as a location map, and write the extension prompt from a ready-made template. Then you'll set up the generation, link the tags and join the two parts in an editor.

Important: Seedance restricts the use of real people in reference videos, so this technique works with generated, animated or anthropomorphic characters.

All prompts used in this lecture are in the attached resource file.

The Seedance 2.5 Editor is a tool for precise video editing. It lets you replace, remove or add a specific object in the frame without affecting anything else.

In this lecture we make three edits to a single clip. First, two tipper trucks are replaced with walking elephants. Then a fruit basket is removed together with all the scattered fruit. Finally, a regular car is replaced with a police car from a reference image, keeping the same trajectory and even the dust from under the wheels.

Along the way you'll get to know the selection tools - rectangle, arrow, brush, point and eraser - and learn how to write an edit so that it works: in your own words, short and specific, always ending with the condition to keep the rest of the frames unchanged.

The example prompts from this lecture are in the attached resource file.