In DOTV a subject is one set of reference material for the same person, object or place: a main picture, a few other views, plus a short description, and optionally a character video and a bound voice. Once it is in the subject library, the next film with that person or object starts by adding it to the canvas instead of picking the reference pictures all over again. This section covers creating a subject, finding and editing what is in the library, and putting a subject to work.
Three ways in:
Inside the dialog, top to bottom:
| Field | What goes in it |
|---|---|
| Main picture | The baseline frame of the set. This is what image and video generation references. |
| Add other views | At most five (with a character video this can be empty). Upload them yourself, or generate them with Smart fill below. |
| Character video | Optional, one continuous shot of 3–8 seconds. A clip you pick that is outside 3–8 seconds is refused on the spot; one brought in from right-clicking a video node outside that range only lends its cover frame as the main picture. When the main picture is still empty, uploading a video also grabs a frame into it. Adding the subject to the canvas places one more node for this video (see Putting it to work). |
| Subject name | Up to 20 characters; the counter on the right is that limit. |
| Voice | Optional, no voice by default. Pick one from our voice library — you can preview it, and you can pick a voice you cloned yourself. Adding the subject to the canvas places one more audio draft node that already has this voice selected (see Putting it to work). |
| Tag | Characters / Scenes / Props / Effects / Other. Optional — left unset, the subject is filed under Other, which is also what the library's tabs filter by. |
| Subject description | Up to 100 characters on what the subject looks like. This text travels with the subject into the video prompt (see Putting it to work), so write how it looks, not what happens in this shot. |
The × at the top right and Cancel: if anything was changed they ask "Exit? Closing will not save any changes." first; with nothing changed they close straight away. Create stays grey until there is a main picture, at least one other view or a character video, a name and a description.
The pill inside the dashed cell generates other views from the main picture: it fills whichever of the three standard views — right, left and back — are still missing, each landing in its own slot as it arrives. Pictures you uploaded yourself do not count as standard views, so uploading a few does not stop it from filling all three; once the three are there the pill disappears (the dashed cell stays, so you can keep adding your own up to five). A set holds at most five views, so when there is not enough room it fills as many as fit. The prompts follow the Tag you picked — characters and props get a side view and a back view, a scene gets another camera position in the same place. With no main picture the pill is grey; hovering it says to add the main picture first.
The pill carries what this run will cost in credits (counted from the number of views it is about to generate). What comes back is drawn in the same style as the main picture; anything that fails leaves its slot empty, and clicking again fills only what is missing.
The pill at the bottom right of the description field reads the main picture and writes a description of up to 100 characters. It replaces the whole text already in the box (the toast afterwards says so), describes only what is visible and invents no brand, person or place names. It carries its own cost on the pill too. With no main picture it is grey as well.
The row in the title of the Assets dialog switches between Assets and Subject library. This screen is a search box, a row of tabs (All / Characters / Scenes / Props / Effects / Other) and a grid of cards. A card carries the main picture, the name, and its tag with the number of views.
The pictures of a subject count against your cloud storage: a subject uses the picture already on the node (or the one uploaded), and the library only remembers which ones belong together. So deleting a subject never deletes a picture, nodes using the same picture are unaffected, and whatever nothing uses any more is reclaimed by cloud storage cleanup (see 14.5).
Add to canvas on a card drops the whole set onto the canvas: the main picture and every other view become image nodes in a single row. If the subject carries a character video, a video node joins the end of that row (same row, same spacing). The description becomes a text node on its own row below. If the subject has a voice bound, an audio draft node lands to the right of the description — it already has that voice selected, so the lines you write are spoken in it. That audio node is not linked: a draft has no sound to reference yet. Once it generates audio, you can connect that audio to a video draft as a reference audio.
At the same moment, if a node with a composer is selected (image, video, text, script, audio, Smart edit and the 3D previz stage all have one), those nodes are linked to it as references — for the next image or video generation they are the pictures and the text to reference, and the character video goes in as a reference video (only models that take reference video actually use it; for the rest it just sits on the canvas). You can name them in the prompt by typing @ (see 10.5). Not every node takes them all: what it cannot take is placed but not linked, and the toast afterwards says how many made it in. Nor is the number of references unlimited — each video model takes a different amount, whatever does not fit is placed but not linked again, and in the composer's reference row the ones the model dropped fade out (see 4.2). With nothing selected the nodes are simply placed.
The description text node joins the video prompt automatically — a connected text node is already used as reference prose, so there is nothing to copy across by hand.
Afterwards the draft that was selected is still selected and the composer does not collapse — the references that just came in are sitting in its reference row, ready for the next prompt.