LESSON 6 · MODES

Three input modes: first frame / first and last / reference

Seedance 2.0 accepts more than just text — you can feed the model images and video as visual input. But there are three fundamentally different ways to "feed it an image," and the model reads each one differently. In this lesson we break down how a "first frame" differs from a "reference" and when to choose each mode.

Why three modes at all

The model can interpret one and the same image in completely different ways: as a literal start of the clip, as a point of departure for motion toward a finale, or as a source of traits to be adopted. The mode you pick determines exactly what you'll see in the result. Let's go through all three.

1
First frame
image → video: the picture literally becomes the first frame
2
First and last
interpolation: you set the start and finale, the model fills in the motion between them
3
Multimodal reference
up to 9 images + 3 videos + 3 audio as sources of traits, not as frames

Mode 1. First frame (image → video)

You feed one image, and it becomes the literal first frame of the clip. The model doesn't redraw the picture — it takes it as is and continues the motion from it: your starting frame comes to life. This is the most predictable way to open a video on a specific composition.

  • What you feed. One image — the exact start of the scene.
  • What the model does. Keeps the first frame identical to your picture and unfolds motion from it according to the prompt text.
  • Where its strength lies. Full control over how the clip begins — pose, angle, composition are all set by you.

Mode 2. First and last frame (interpolation)

You set the start and finale frames, and the model builds a smooth motion between them. Handy when not only the beginning matters but also how the clip ends — for example, when you need to get from point A to point B along a specific trajectory.

  • What you feed. Two images: the first frame and the last frame.
  • What the model does. Builds the transition — a morph, camera move, or object movement — so that the clip begins with the first frame and ends with the second.
  • Where its strength lies. Control over the clip's arc: a guaranteed finale, a defined "where we're heading."

Mode 3. Multimodal reference

Here images and video work not as frames but as sources of traits. You can feed up to 9 images, up to 3 video fragments, and up to 3 audio clips, and each reference is responsible for its own thing: character, scene, style, motion, rhythm. You point to references directly in the prompt text with tags — [Image 1], [Video 1], [Audio 1].

  • Images. Up to 9 — the hero's appearance, setting, palette, style.
  • Video. Up to 3 (no longer than 15 seconds in total) — timing, gesture rhythm, camera movement.
  • Audio. Up to 3 (no longer than 15 seconds) — the rhythm and soundtrack the scene is built around.

Image references are stronger at spatial things — appearance, mood, style. Video references are stronger at temporal ones — timing and motion. Audio sets the rhythm. Combine them to fit the task.

Comparing the three modes

ModeWhat you feedWhen to use
First frameOne image (= literally the first frame)When you need a precise start of the clip that you define, and its coming to life
First and lastTwo images: start and finaleWhen the final frame matters too — you need a defined transition from A to B
Multimodal referenceUp to 9 images + 3 videos + 3 audio (sources of traits)When you need to adopt appearance, style, motion, or rhythm rather than lock down a frame
🎬 What this looks like in practice
Take a photo of a person sitting by a window. In "first frame" mode this image becomes the literal starting frame — and comes to life: the model builds in natural motion, the person slowly turns their head, the curtain by the window sways in a light draft, the light shifts a little. The composition, pose, and angle stay exactly as in your picture, because this is precisely the first frame of the clip.
⚠️ A common mix-up: reference ≠ first frame
Even if in multimodal reference mode you feed just one image — it does NOT become the first frame. The model treats it as a reference: a source of traits (the hero's appearance, style, palette) that it adopts, while building the composition and start of the clip anew from the prompt text. If you want the picture to literally open the clip and come to life — choose "first frame" mode. If you want to adopt traits without being tied to a specific frame — choose "reference."

How to choose a mode for your task

You need a precise start

You have a ready composition, and it should open the clip unchanged — go with "first frame."

You need the finale too

It matters how the clip ends and what path it travels from beginning to end — go with "first and last."

You need traits

You want the same character type, style, or motion but in a new scene — go with "multimodal reference."

✅ Tip
Before generating, ask yourself: "Do I want this image to be literally the first frame?" If yes — that's "first frame" mode. If the image merely hints to the model HOW something should look or move — that's reference mode. The same file in different modes will give a completely different result.

In the next lessons we'll cover how to precisely address references with the tags [Image 1] / [Video 1] in the prompt text and distribute roles across several sources.

Try Seedance 2.0 yourself

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

Start Creating