Prompting: how to talk so pod hears you
A handful of habits — front-loading your brief, naming outcomes instead of mechanisms, and scoping your notes — turn pod from a machine you argue with into a crew that reads your mind.
pod understands plain language, in any language. There is no secret syntax to learn. But like a real crew, it works best when you brief it the way a director briefs a set: say what you want the audience to feel, say it early, and be precise about which shot you mean. This page is the craft of that. (For how conversation works mechanically — gates, the front desk, what Enter does — see talking to pod.)
Front-load the brief
Before anything starts, pod asks for what it needs — duration, language, look, dialogue style, shape (16:9 wide, 9:16 vertical, 1:1 square). It only asks about the blanks. So the fastest sessions start with a first message that fills the blanks in advance:
| Before | After |
|---|---|
| "make a teaser of my script" → pod has to ask: how long? what language? wide or vertical? what look? |
"90-second telugu teaser of the attached script, vertical, moody like a Rajamouli intro, protect the ending" → zero questions; straight to the script gate |
Take that "after" apart — it's the anatomy of a good brief:
- "90-second" — duration, stated, not asked.
- "telugu" — language. This locks dialogue and casts the right voices.
- "teaser of the attached script" — what it is, and from what.
- "vertical" — the shape (9:16, reels-ready).
- "moody like a Rajamouli intro" — the look, as an outcome (more below).
- "protect the ending" — a standing instruction: tease, never reveal. pod respects this through every shot.
Outcome, not mechanism
You don't have to know how the images get made. Say what it should feel like and let the crew translate. "Moody, like a Rajamouli intro" works better than trying to guess technical settings — pod's cinematographer knows what that means and carries it into every prompt. When you do want to speak precisely, pod has a full craft vocabulary — shots, lenses, light, grades — on the camera language page.
Style vs direction
Two different dials, and pod treats them separately:
- Style = the visual look: "anime", "35mm documentary", "noir monochrome".
- Direction = tone, pacing, voice: "slower, let moments breathe", "funnier", "more menace in the villain's lines".
You can set either in the brief (--style "…" and --direction "…" exist if you like flags, but plain words do the same job). If a scene looks right but feels wrong, you want direction, not style.
Scoped notes vs standing rules
At every gate, pod listens for how big your note is:
- Scoped — name the shot, say one change: "shot-05: make it a POV". That changes one shot and nothing else. This is the single most useful habit at the keyframe and clip gates: name the shot ("shot-07"), say ONE change.
- Standing — say "from now on": "from now on, illustration style". pod remembers it for the rest of the episode. In series mode, standing rules persist across the whole series — say it once in episode 1 and it still applies in episode 6.
Brief the critics too
In assisted and autopilot modes, AI critics inspect every image and clip. They judge against your intent — so tell them what your intent is. Two directions worth giving:
- Relax something deliberate: "the clock being wrong is intentional" — otherwise a perfectly good take might get flagged for the very detail you designed.
- Tighten something you care about: "be strict about faces".
Teasers protect endings — if you say so
Tell pod "tease it, never reveal the ending" (or just "protect the ending" in the brief) and it will build the whole teaser around withholding. And a house note on voice: in trailers and teasers, characters speak in their own voices — pod won't put a narrator over your film unless you ask for one.
Choosing how people speak
Three dialogue styles, asked upfront (or set with --dialogue):
| Choice | What it means | When it fits |
|---|---|---|
| native | Lines performed on camera — lips move with the words | Screenplays with real scenes and dialogue (this is the house default for screenplays) |
| vo | Voice-over narration laid over the footage | Explainers, documentary formats, narrated stories (the default for narration formats) |
| silent | No speech at all | Mood pieces, music-driven teasers, ambient work |
None of these are final: after delivery you can dub a real recording over any shot — see your own media.
"Assume" — the skip button
Anything you leave open, pod lists as an assumption and waits for your explicit OK — nothing renders until you say go. If you'd rather not review that list, say "assume" at the question, or put "you're ok to assume" right in the brief, and pod fills the blanks itself. After a couple of productions it gets even easier: pod remembers your habits and offers "your usual" (telugu · 9:16) — still confirmed, never applied silently.
"No" is always free
A bare "no" at any gate never spends money. pod stops and asks what to change. You cannot decline your way into a bill — the only thing that spends credits is footage you approved at the cost gate.
A brief in the wild
❯ 90-second telugu teaser of the attached script, vertical, moody like a Rajamouli intro, protect the ending. you're ok to assume the rest
Got it — 90s · telugu · 9:16 · teaser · ending protected.
Assuming: native dialogue (it's a screenplay) · 1080p delivery.
Say go and I'll start on the script.
(one message, zero follow-up questions)