Generating images

AI video creation usually starts with images: storyboard frames, scene shots and character designs all begin as an image node. On the DOTV canvas every image comes the same way: create an image node, write a line in the composer underneath it, pick a model, hit send. This section covers what each control on that card is; the sections after it cover what you can do with an image once you have one. A freshly created image node is a draft: the picture is empty, and selecting it floats the composer up underneath (see 2.1 for how to create nodes).

What is on the composer

Quality only matters for gpt-image-2.5 (the default Gemini image model ignores it, and pricing is flat per image either way). Style opens the style library: everything in there was published by someone else (or by you). Pick one and the chip folds away, replaced by a cover thumbnail on the row below (with Style written across it) — hover it and it reads Replace, click to pick another, and the × at its top right takes it off; when you generate, that style's description is appended to your line while settings like the model and the ratio are swapped over too. If the one you picked has a cover, a style node is also dropped to the left of this node and wired in — so you can see at a glance that this image came from that style, and you can drag it onto other nodes; cut that wire or delete that node and the style comes off with it. Camera is a gear set (body / lens / focal length / aperture); leave it on and that gear phrase is appended to the prompt. Marker opens element selection, where you circle the spot you mean on an image that has already rendered (the same thing as on the video input bar, see 4.2). Camera and presets are mutually exclusive: with the camera on, Preset is grayed out; with a preset applied, the Style and Marker entries are not shown at all.

The composer at the bottomWrite the prompt in the middle. The row below runs, left to right: the model, the specs (Ratio / Quality / Sharpness / How many), Preset and Camera.

The prompt goes in the middle; the row below is, left to right, Model, Spec, Preset, Camera.

Prompt — plain English is fine. You can use @ inside it to refer to other nodes on the canvas.
Reference — the + Reference chip above puts you in "pick a reference from the canvas" mode: click an image node on the canvas to wire it up. Whatever is wired shows as thumbnails on the row below, and double-clicking one moves the canvas over to it. With a reference in play you are generating from an image, not from text alone.
Model — eight options: Gemini Image (Nano Banana 2) (the default), Gemini Image Pro (Nano Banana Pro), Seedream 4.5 Image, FLUX.2 Pro Image, Seedream 5.0 Pro / Lite Image, Midjourney and GPT Image 2.5. Midjourney creates one image per run. A few presets name a model of their own (see 3.5).
Spec — the button reading "16:9 · High · 2K · 1 image". It opens Quality, Sharpness, Ratio (ten of them, 16:9 by default) and How many (1 to 4 images). Change the model and this button holds something else: Midjourney leaves only the ratio (it has no quality, sharpness or count tiers). If an older node still uses Qwen Multi-Angle, it shows the three camera groups — Orbit (eight positions around the subject), Tilt (-30° looking up / 0° eye level / 30° / 60° looking down) and Shot scale (close-up / medium / wide), one image from one position per run.
Preset — fifteen ready-made recipes (see 3.6).
⚡ and send — the number at the bottom right is what this run will cost in credits, and the round button beside it sends. While it is generating, that button becomes stop.

While it runs, and after

Once you submit, the node goes to generating: a progress skeleton where the picture goes, with cancel in the middle.
When the image is done, the top right of the node title shows its actual resolution, such as 1920 × 1080.
When one run makes several images, the first one fills in this node and the rest go into History in the bottom toolbar, ready to be pulled back onto the canvas at any time.

Another take

Every generation costs credits, and cancelling only stops the run that has not finished yet. More images and higher sharpness both cost more, so glance at the number on the ⚡ before you send.

Select a node that already has an image and the composer is still there. Change the prompt, send again, and you get another take:

Change the Ratio and a draft node reshapes on the spot; a node that already has an image only reshapes once you submit.
The new image replaces the one that was on the node. To keep both, press ⌘D on the node for a copy first.

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

Open DOTV Download the desktop app

3.1 Image nodes

3.3 Editing the picture

Generating images

#What is on the composer

#While it runs, and after

#Another take

What is on the composer

While it runs, and after

Another take