AI Guides › Workbench
By Nigel Guy · 8 min read
The usual mistake with a "documentary from one prompt" tool is believing the "one prompt" bit. You type a topic and a mood, approve whatever comes back, and pay for 20 minutes of narration over footage that is roughly on-topic. Then you spend the evening fixing what a tighter brief would have avoided. VidRush does handle the research, script, voiceover, music, footage and render. It just can't guess the angle, the facts you care about or the shots you need, and any gap you leave gets filled with generic material.
The rule: write the brief as if you were briefing a human researcher. Check the Quote Statement before you approve it. Then fix the result in small, specific edits, never one big "make it better".
All figures come from VidRush's own pricing page and docs at time of writing. VidRush bills in US dollars, so the pound figures are rough conversions and will move with the exchange rate.
| Item | What it does | Cost at time of writing | Best for | Catch |
|---|---|---|---|---|
| Deep Video v1 Mini | Images plus motion graphics | Cheapest model; the docs list 25.6–32 credits per minute depending on reasoning | Topics with little usable footage, and test runs | No video B-roll |
| Deep Video v1 Pro | Video B-roll plus images | The docs list 35.2–44 credits per minute | History and events with archive footage | Costs more, and it only helps if footage exists |
| Plans | Monthly credit allowance | Starter $99 (about £74), Creator $239 (about £178), Pro $639 (about £477) a month; 20% off on annual billing | Regular publishing | No free trial is listed; unused credits roll over only up to a cap |
| Quote Statement | Shows VidRush's reading of your request before it generates | Free | Catching the wrong length, format or title | It's easy to click "Approve & Continue" without reading it |
| Rush Agent | Edits in plain language inside the editor | Each action is billed as a small fraction of a credit | Swapping clips and fixing short stretches of voiceover | Early beta; best kept to about 30 seconds of video per request |
| Re-render | Re-exports after edits | About 10% of the original generation cost | Final export | Every round of edits costs credits |
VidRush's pages disagree slightly on per-minute rates (roughly $1.56–$2.72 a minute), so budget from the credit figure in your own Quote Statement.
VidRush's own guide organises a prompt into four parts: a director's brief (the angle), a style guide, a target audience and key facts. The template below uses that structure and adds the parts a good prompt needs: what a good result looks like, guardrails, and a self-check. Fill in every square bracket and delete any line you can't fill honestly.
Title: [WORKING_TITLE]
Format and length: a [6–8 / 10–12 / 18–20 / 30–40]-minute documentary for YouTube.
Director's brief (the angle, in 2–3 sentences):
This film explains [SUBJECT] by following [THE ARC — e.g. origin, peak, decline]. The question it answers for the viewer is: [CENTRAL_QUESTION]. It should end on [CLOSING_IDEA].
Audience: [WHO_WATCHES — e.g. curious adults with no specialist knowledge], who usually watch [SIMILAR_CHANNELS_OR_TOPICS].
Style: [TONE — e.g. calm, curious, unhurried]. Short spoken sentences. No hype, no cliffhanger teasers, no rhetorical questions in a row. Open with: "[EXACT_HOOK_LINE, or delete this line]".
Structure — cover these talking points in this order:
1. [POINT_1]
2. [POINT_2]
3. [POINT_3]
4. [POINT_4]
5. [POINT_5]
(Use about one point for every two minutes of runtime.)
Key facts to use exactly as written (do not round, restate or embellish):
- [DATE / NAME / FIGURE 1]
- [DATE / NAME / FIGURE 2]
- [DATE / NAME / FIGURE 3]
Sources to research from: [URL_1], [URL_2]
Visuals: prefer real archive footage of [SPECIFIC_SUBJECTS]. Where no footage exists, use stills or simple motion graphics rather than generated scenes that look real.
Guardrails: do not invent quotes, statistics or eyewitness accounts. If a fact cannot be confirmed from the sources, leave it out rather than guess. Handle [SENSITIVE_PART — e.g. accidents, deaths] factually and briefly, without dramatisation.
Before generating, confirm that every talking point has visual material available, and flag any point that does not.
What to fill in: the title, the length band, your angle, audience, tone, talking points, the facts you have verified yourself, your source links, and any sensitive section that needs careful handling.
| Part | Job | If you leave it out |
|---|---|---|
| Title line | VidRush builds the thumbnail from the title | You get a generic title and a thumbnail to match |
| Format and length | Sets the length band and how much it charges | VidRush guesses, and you correct it in the Quote Statement |
| Director's brief | Sets the angle | You get a Wikipedia-order summary |
| Audience and style | Controls pacing and register | It defaults to "YouTube voice" |
| Ordered talking points | Gives the structure, at about 0.5 points per minute according to VidRush's guide | Sections get uneven weight |
| Key facts | Reduces improvisation | Dates and figures drift |
| Guardrails | Stops invented colour | Made-up quotes are the worst failure, and the hardest to spot |
| Self-check | Uses VidRush's Input Validation step | You find footage gaps after you've paid |
If you're testing a topic, the minimum is the format and length line, a two-sentence angle, five ordered talking points and the guardrail line. That comes to about 600 characters, well over the 200-character minimum. Test it on Mini at the shortest length. If the structure holds up, spend Pro credits on the full version. Change one thing per attempt, starting with the director's brief, as VidRush's guide suggests.
Rush Agent works best on targeted edits of about 30 seconds of video, so give it a list of separate, specific requests rather than one paragraph of changes. Paste it into the agent one numbered item at a time.
You are editing my VidRush project "[PROJECT_TITLE]". Make only the change below, then stop and tell me what you changed and the new timestamps.
Change: [ONE_SPECIFIC_EDIT]
Location: [TIMESTAMP_START]–[TIMESTAMP_END] (keep this under about 30 seconds)
Keep unchanged: narration outside this range, music, and the overall runtime unless I say otherwise.
If the change needs footage you can't find, describe the closest real footage you can source and ask me before inserting anything generated.
Example items for the [ONE_SPECIFIC_EDIT] slot, one per request:
Restructuring a whole section, for example cutting one segment to a minute and spending the saved time on another, is too big a request for the agent while it's in beta. Do that in the brief and regenerate, or do it by hand in the editor's timeline.