> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wji.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Video generation

> Duration, first and last frames, task modes, reference video and audio, and upscaling.

Switch to video mode with the icons at the top of the form. Most of the form is the same; this page
covers what is different.

## Duration

A slider whose range comes from the model. Switching models applies that model's default.

Seedance models offer **auto duration**. With it on, the slider disappears and the model decides how
long the clip should be. The cost estimate then shows a per-second rate rather than a total, because
the final charge depends on what the model produces.

## First and last frames

<Frame caption="First and last frame slots">
  <img src="https://mintcdn.com/wji/1nwL1jaWRdmAhYLq/images/image-generation/frame-slots-01.png?fit=max&auto=format&n=1nwL1jaWRdmAhYLq&q=85&s=83c3f5b7505647fccdc9db0f4ba20a00" alt="First and last frame slots both filled, with the swap control between them" width="1600" height="900" data-path="images/image-generation/frame-slots-01.png" />
</Frame>

If a model supports them, slots appear for a start image and an end image. You can upload a file,
pick from your asset library, or take a frame from a history result. A swap button exchanges the
two.

Filter the video model list by *first frame* or *first and last frame* to see only models that
support this.

## Task modes

Seedance models offer three:

| Mode         | What it does                           | Requires                                                        |
| ------------ | -------------------------------------- | --------------------------------------------------------------- |
| **Generate** | Text to video, or multimodal reference | Nothing extra                                                   |
| **Edit**     | Modify an existing clip                | At least one reference video, and edit keywords in the prompt   |
| **Extend**   | Continue an existing clip              | At least one reference video, and extend keywords in the prompt |

<Warning>
  Task mode and prompt wording must agree. Edit mode needs words like *edit*, *add*, *remove*, or
  *replace* in the prompt; extend mode needs *extend* or *continue*. A generate-mode prompt
  containing edit keywords will be rejected so it is not misclassified.
</Warning>

Edit and extend mode force the aspect ratio to adaptive and hide the ratio picker. Edit mode also
forces MOV output.

## Video upscaling

A separate capability with its own settings:

* **Scenario** — general, UGC short video, AIGC, short drama, or classic film restoration
* **Enhancement level** — fast, standard, or pro
* **Resolution** — original, 720p, 1080p, 2K, or 4K
* **Bit depth** — 8, 10, or 12-bit

Seedance results also offer HD upscaling directly from the history card, provided the original has a
seed. The available resolutions depend on the model and the original resolution.

## Text to audio

The fourth mode generates speech. Its settings come from the TTS model: speaker, speech rate, pitch,
loudness, sample rate, and format.

## What video editing is not

WJI has no timeline editor in this app, and no way to cancel a running generation. Selective
regeneration of part of a clip exists, but only on the
[Infinite Canvas](/infinite-canvas/video-tools).