Guide
English
Search the docs⌘K
6.5

Preparing assets

The same character has to have the same face across a dozen shots — the hardest thing in AI video creation. DOTV's way to get it is to generate one reference portrait of them first, then carry that picture along as a reference every time a shot is generated; that is what preparing assets means. An "asset" is anything in this film that comes back again and again and has to look the same. There are three kinds:

  • Characters — people. The same person in two looks counts as two assets ("Lin An, present day" and "Lin An, period costume" get one each), or the model will not know which version to draw.
  • Scenes — places. The overpass, the tunnel, the convenience store.
  • Props — objects that keep appearing. Those running shoes, that cup.

This screen

Three groups of asset cardsCharacters / Scenes / Props, three groups. Each card is one asset: a name, a line of description, and its reference art. An empty slot means the art isn't generated yet.
1 / 2
Characters, scenes and props in three groups, one card per asset.

When the script is generated, the assets get listed for you. Each card carries a name, a line of description, and the reference art (an empty slot until it has been generated).

Every card can:

  • Generate — produce one from the prompt on the card. What comes out lands on the canvas as an image node wired to the script node, so changing it or swapping it works like any other image on the canvas.
  • Upload — if you already have art (a character turnaround from the client, a photo of the real location), send it straight up.
  • Clear image / Delete — the first drops just the picture and leaves the card to be regenerated; the second removes the asset itself.

Generate all assets at the bottom lets you tick a batch and do them in one go; next to it is a count of how many are still missing, which turns into "All assets are ready" once everything is in — and the button turns into "Next: compose prompts →".

Global visual style

Three groups of asset cardsCharacters / Scenes / Props, three groups. Each card is one asset: a name, a line of description, and its reference art. An empty slot means the art isn't generated yet.
1 / 2
Assets are listed first when the script is generated; the row along the top is the look for the whole film.

The group at the top is the Global visual style, and it is where the film's consistent look comes from. It has two halves:

  • Style reference images — a row of same-sized cards, one image per card: when asset art, storyboard images and video frames are generated, these are sent to the model along with everything else, and it takes the style only, not the content (the prompt says so to begin with). The last cell is the way in to upload — click it to pick a file, several at a time if you like; you can also pick nodes on the canvas that have already generated. 4 at most — each one uses up one of the reference slots for image generation, so the more you pick, the fewer slots are left for the content of the picture.
  • The style description — "night city · cinematic · cool tones", for instance. It is appended after every asset image and every storyboard prompt, and the two are one thing said two ways: the images set the tone, the words add the detail.

To change the feel of the whole film (from "cinematic, cool" to "warm film stock"), swapping the reference images and editing this one line saves a great deal of work over going shot by shot. Click elsewhere and it saves itself. Remove on a reference card only stops using it — the node on the canvas is not deleted, since it may well have been a picture on the canvas to begin with.

Give a character a voice

Character cards also let you pick a Voice. One is assigned by the model when the script is generated, and you can listen and change it. When dialogue is generated in bulk, each character's lines use their own voice — which is why the lines are bound to characters back in 6.3.

Try it in DOTVReading only goes so far: open DOTV and make an AI video with the feature this section covers.