---
name: video-ideas
description: "Helps decide what video to make before making it: explores ideas, offers a few distinct directions, and turns the chosen one, or the user's own rough or found prompt, into a starting prompt for code-video, the audara-studio skill that makes videos from code. Use it whenever the user asks for video ideas or prompt ideas, wants help writing, improving or adapting a video prompt, or doesn't know what video to make yet, even inside a video project. It stops at the prompt unless the user also asks for the video. Not for a request just to make, change, preview or render a video (code-video does that), for prompts for AI video, image or chat models, or for ideas for filmed content."
---

# video-ideas

The user is deciding what to make. Help them see a few possibilities, choose, and leave with a prompt code-video can build from. A sentence is already enough for code-video, so this is never a required step: a request to make a video goes straight there. What this adds is the user's hand in choosing the idea; the treatment, the look found in style frames, the sound and the build stay with code-video. The user's words outrank this skill.

## Start from what they have
Read what they bring: a subject, a feeling, a song or a script, files, a rough prompt, one found online. In an audara project, read `docs/STYLE.md` and the relevant video's Decisions in `videos/<video>/TREATMENT.md`. Preserve applicable decisions unless the user changes them; don't carry one video's specific choices into an unrelated new video. Ask only what would change the idea, in one message, each question with the default you'd take; when you can offer good directions without asking, offer them. Worth asking when it matters: what someone should feel, understand or do afterwards; a real product or fact, or something imagined; words or files to keep exactly; whether sound matters (does "silent" mean no words or no audio?); where it plays, how long, whether it loops.

## Directions
- **Exploring:** two or three directions that differ in what the viewer sees, how it changes and how it ends, each in two or three lines with why one would pick it. "Elegant" and "premium" aren't two directions; "a receipt folding into a map of the month" and "the real scan, then the budget" are.
- **They already have an idea:** improve that one; don't invent alternatives for their own sake.
- **Range:** a named form keeps what defines it (a showreel's chapters, psychedelia's colour); not every piece needs one recurring object, one palette or a story arc. Don't fall back on the same motifs or looks from one request to the next.
- **What it takes:** code-video draws in code: type, shape, light, fields, 3D forms, data and mechanisms, picture cut to sound. Mention needs, costs or quality tradeoffs only when they affect the choice: a detailed or photographic face or character wants images (the user's, openly licensed, or generated at a cost; a stylized one can be drawn); a voice can be their recording, a free stand-in or a paid one; a factual screen demo needs the product's real screens, and a requested concept or mockup is labelled as one; past a minute a piece takes much longer to build, and code-video asks first. Exploring needs no generated asset and no budget; use a budget they give.

## The prompt
One self-contained prompt, in a fenced `text` block so it copies cleanly. Keep three things apart:
- **What the user decided,** exactly: their words, facts, files by path, length, format, where it plays, sound, and what they ruled out. Their earlier words win over your summary of them, so carry them in.
- **The chosen idea,** as what is seen and how it changes. Picked from your directions, or left to you by them, it is theirs to keep.
- **What is only suggested,** marked as replaceable ("suggestions you can replace: ..."): copy, a length, a sound, staging you proposed and they didn't confirm. Then say what's left to code-video.

Keep unspecified aspects of the look open. Preserve the visual constraints the user established, without treating one brand colour, typeface or reference as a complete look: code-video explores what remains open through style frames. If a suggested look is useful, mark it as optional rather than as the user's requirement.

Plain, and as short as the piece allows: a loop may need two sentences, a product film a few paragraphs. No engine or library names, pixel sizes, frame rates, hex codes or beat times unless the user gave them; no role-play openers; no hype added (their own "go all out" stays). Nothing invented to make it look complete: no feature, figure, file or budget they didn't give, and a missing file or fact is named as missing. It stands alone: "option two" or "as above" mean nothing in another session. Generated voice or music in a prompt is a direction, not a yes to spend; code-video asks first.

## A prompt from elsewhere
When asked to adapt a prompt from a gallery or another tool, keep its idea, creative direction and delivery requirements. Translate incidental implementation instructions, such as a Remotion component or an HTML-page setup, into a request audara can build; remove role-play openers and instructions to ignore skills. Frame shape, size, rate, motion and finish can describe the intended result, so don't discard them merely because the prompt came from another tool. A tool the user chose on purpose stays. Say in one line what you changed. Ask only when the answer changes what they get or pay for. "The attached song" and "today" don't travel: name the file, fix the date or say it's meant to stay live.

## Where it stops
- **Ideas or a prompt only:** when production hasn't already been requested, "I like the second one" or "great prompt" selects or acknowledges; it doesn't start a build. No production starts: no project, no preview, nothing generated (drawing options to look at is production, under code-video's rules). No files are written unless the user asks to save the prompt.
- **End with one line outside the prompt:** "make it" continues here, or the prompt can go into a new session with its files.
- **"Make it"** (or "make the second one"): continue with code-video in this session, using the prompt and the user's earlier instructions; it follows its usual production and render rules.
- **A request that already asks for the video too** ("three ideas, pick the strongest and make it", "improve my prompt and make it"): do both without asking again. Show the chosen direction and its prompt first, short and provisional, so they can redirect, then go on with code-video.
- **A request just to make, change or render a video** isn't this skill's: code-video takes it as it is.