Shot devices: how one scene is filmed
The studio has three layers and they do not replace each other.
| Layer | Answers | Where it lives |
|---|---|---|
| Style | what the whole clip looks like | Prompts section, Styles tab |
| Story frame | what we are filming at all | same section, Frames tab |
| Shot device | how one scene is filmed | Devices tab |
A style and a story frame are set once per track. A device lands on one scene card and touches nothing else in the project. That is why there are fifty-six devices and only fifteen styles.
A device is the difference between two frames
A scene here has two image prompts: prompt (the first frame) and last frame prompt. A paid engine builds the video between them. So motion is not defined by a camera path — it is defined by how the last frame differs from the first.
This is worth spelling out, because every other prompt bank on the internet works the other way. There a device is a line like "drone orbit, 8 metre radius, 84 degrees". That line does not fit our frame pair at all: we do not have a minute of flight, we have "here is the start, here is the end, get there".
So a push-in looks like this here:
- first frame: medium shot, character from the waist up, air above the head;
- last frame: same character, same light, same clothes — but only head and
shoulders, filling two thirds of the frame;
- motion: camera travels on one axis, subject stays put.
Three fields instead of one trajectory. In exchange it works on any engine that takes a first and last frame, and does not depend on whether the model knows the word "dolly".
There is no style inside a device
The device texts contain no words about film, anime, grade or grain. That is deliberate: the pipeline supplies the style, and if the device supplied one too, the prompt would carry two different worlds and the model would pick between them on its own.
Practical consequence: the same device works identically on Ghibli and on VHS. You can change the style of the whole track without touching a single scene.
What Apply actually does
A device card has slots: {character}, {location}, {time of day} and so on. Fill them, press Apply, and five fields land on the scene card at once — first frame prompt, last frame prompt, motion prompt, shot size and camera move.
This is free. A device is text, and text costs no credits in this studio. You only pay when you press Generate frames.
An unfilled slot substitutes the example from the dictionary. That is better than leaving {character} in the prompt: models read curly braces literally and draw the braces.
The six shots every clip is built from
Below is the pack that applies in one click. It exists to cure the single most common storyboard disease.
The disease: all thirty scenes are medium shots. The clip reads flat and you cannot see why by eye — every individual frame is fine.
The cure: five different shot sizes across the first six scenes.
| # | Device | Shot size | What it does |
|---|---|---|---|
| 1 | City establishing shot | establishing | says where we are, once |
| 2 | Push in on the face | medium → close-up | a moment of attention |
| 3 | Cut-in detail | extreme close-up | punctuation |
| 4 | Locked-off frame | wide | somewhere to breathe |
| 5 | Walking alongside | medium | the sense the track is going somewhere |
| 6 | Pull back to reveal | close-up → wide | scale in a single scene |
All six are open on the free plan.
Steps
- Build the storyboard as usual — Generate storyboard.
- Open the lesson and press Apply pack. It lands on the first six scenes.
- Fill three slots once for the whole pack: character, location, **time of
day**. The same character in all six — otherwise the pack does not assemble into a clip, it scatters into six postcards.
- Walk the strip and fix the text where a device is not about that lyric. A device
is a starting point, not a verdict.
- Only now press Generate frames.
What to look at
Open the contact sheet after the frames are generated. Six thumbnails in a row should be obviously different sizes: city, face, fingernail, room, stride, whole room. If everything on the sheet looks equally close, the devices did not land — check that you pressed Save frame before regenerating.
The usual mistakes
Slots filled differently in different scenes. "Guy in a hoodie" in scene one and "the hero" in scene four are two different people to the model. Fill the pack once.
Applying the pack after the frames are already drawn. The text is replaced, the pictures are not, and it looks like the button is broken. Apply before generating, or press Regenerate frames straight after — that is free.
Putting a push-in on the cheap engine. Grok animates only the first frame: it has nowhere to put the difference between the two. The device card says so — "needs an engine with first and last frame". Devices without that mark work on Grok.
Taking all six and stopping there. The pack covers six scenes out of thirty. After that it is the same five shot sizes on rotation, by hand.
The artefact
Pack "Six shots every clip is built from" (basic_shots) — six cards, applied in one click to the first six scenes. Slots: character, location, time of day. Free plan.
Next: Light: five sources and what each one costs — the one decision that changes a frame more than shot size does.