LESSON 4 · SETTINGS

Model parameters: version, resolution, duration, aspect ratio, seed

A prompt is the "what" you want to see. Generation parameters are the "in what form" the model delivers it: how crisp, how long, in what frame shape, and whether the result can be reproduced. In this lesson we'll break down each Seedance 2.0 parameter one by one and learn to combine them for the task at hand: a quick draft or a final shot.

The five parameters you control

No matter which generation tool you run Seedance 2.0 in, the model has the same set of base settings. Think of them as five levers.

1
Version
Seedance 2.0 or 2.0 Fast — quality versus speed
2
Resolution
480p / 720p / 1080p, natively up to 4K
3
Duration
roughly 4–15 seconds
4
Aspect ratio
frame shape for the display platform
5
Seed
reproducibility and variations of the result

1. Model version: 2.0 versus 2.0 Fast

Seedance 2.0 comes in two versions. They understand prompts and references identically — what differs is the balance of quality and speed.

Seedance 2.0

Maximum quality: best detail, more stable motion physics and character identity. Takes longer to compute and usually costs more.

When: final render, complex scenes, photorealism, close-ups of faces and textures.

Seedance 2.0 Fast

Faster and cheaper, with slightly lower quality. Ideal for checking an idea, composition and motion in minimal time.

When: drafts, iterating through prompt variants, checking framing and timing.

✅ Rule for choosing a version
Don't pay in quality while searching for an idea. First run your concept on Fast — you'll see whether the composition is right, whether the rhythm is right. And only once the shot works, switch to the full version for the final.

2. Resolution: 480p / 720p / 1080p and native 4K

Resolution sets the crispness and size of the frame. The higher it is, the more detail — but the slower and more expensive the generation. Choose based on the clip's purpose, not "always maximum."

480p
drafts, quick check of the idea and motion
720p
working balance of quality and speed
1080p
final for most platforms

The model can output natively up to 4K, but not "with a checkbox" — only when you explicitly ask for detail in the prompt: add ultra-sharp detail, crisp texture, high detail texture. Without such a request, cranking up the pixel count is pointless — you won't gain detail, but time and cost will rise.

🔑 Quality = resolution + a request for detail
Resolution alone doesn't paint pores on skin or fibers in fabric. High crispness only pays off paired with words about texture in the prompt. The number in the settings and the words in the text work together.

3. Clip duration

Seedance 2.0 generates clips roughly 4 to 15 seconds long. Duration affects not only the running time but also how many actions you can fit in.

  • Short clip (4–6 sec). One clean action, one accent. Cheaper and more stable — motion "falls apart" less often.
  • Medium (7–10 sec). Enough for an action with development: approach, gesture, reaction.
  • Long (up to 15 sec). Needed for an arc or several shots. Remember the rule from prompt work: one action per shot — don't try to cram five events into one long frame.

4. Aspect ratio: frame shape for the platform

Aspect ratio is the proportions of the frame. Choose it before generation and for the platform where the clip will live: a vertical video in a horizontal feed shows with black bars, and vice versa.

Aspect ratioShapeWhat for
16:9horizontallong horizontal clips, landing pages, presentations, TV format
9:16verticalvertical short videos for smartphones, stories
1:1squarefeed posts, a universal crop for different platforms
4:3classic horizontalretro aesthetic, documentary and vintage look
3:4classic verticalportrait shots, soft vertical framing
21:9cinematic widecinematic scenes, panoramas, a "trailer" look
auto / adaptiveby inputthe model picks the proportions itself to match the uploaded image or video reference

In image-to-video mode or with a video reference, it's convenient to set an adaptive aspect ratio — the model will fit the frame to the source, and there won't be any unwanted cropping or stretching.

5. Seed: controlling randomness

Seed is the number the model starts generation from. It determines the "random" part of the result. By controlling the seed, you turn a lottery into a controllable process.

You fix the seed

With the same prompt and parameters, the model produces the same result. This way you change one parameter (say, resolution) and compare fairly — everything else stayed the same.

You change the seed

With the same prompt you get a different variation of the same idea. Handy for cycling through takes when you like the scene but want to see alternative executions.

🔑 Why this matters in practice
Found a great shot on Fast/480p? Write down its seed. Carry that same seed over to the full version and 1080p — and you'll get the same shot, just in high quality, rather than a new random result.

The workflow: from draft to final

Let's put it all together into one route that saves time and nerves.

1
Draft
Fast + 480p, short duration — searching for composition and motion
2
Lock the seed
liked the shot — memorize the seed
3
Refine
same seed, tweak the prompt and aspect ratio
4
Final
full version + 1080p, a request for detail in the prompt
✅ The key takeaway of the lesson
Don't generate at maximum quality right away. Build on drafts with a fixed seed on Fast/480p — it's quick and cheap. Once the shot works, raise the version to full and the resolution to 1080p while keeping the seed. That way the final is predictable, while higher resolution always takes longer to compute and costs more — there's no need to spend it searching for an idea.

In the following lessons we'll apply these parameters to specific input modes — text, image, and multimodal references.

Try Seedance 2.0 yourself

Generate your first clip in ZenCreator — Seedance 2.0 is available in the Image-to-Video tool.

Start Creating