- Home
- Skills
- Image & Video
- Stage Generate: AI Footage Production
More screenshots
Works with the AI Agents you already use
Stage Generate: AI Footage Production
Generate AI footage without burning the budget: cost confirmation before the first paid call, talking-head and b-roll patterns, keyframe economics, and the lip-sync rule that saves re-rolls.
$5
Stage Generate: AI Footage Production
Example session with this skill installed
I need AI footage for a 45-second vertical product explainer: one presenter opening and closing, four b-roll shots of the product in between. I keep wasting money on re-rolls. Plan the generation run before I spend anything.
- Read your context and instructions
- Compiled the stage generate
GENERATION PLAN — produced by the skill. Nothing generates until you approve the call count.
Budget confirmation: 6 shots, 1 character, 9:16 throughout. Paid calls: 1 presenter reference still + 1 b-roll style still + 6 clips + verification stills = roughly 10–12 calls. That number is the approval point — if the shot list grows, we re-confirm before generating, not after.
Classification and keyframes: shots 1 and 6 are talking head (presenter open and close) — one keyframe each, since only the expression shifts; an end frame would buy nothing. Shots 2 and 4 (product on desk, gentle push-in) are one-keyframe b-roll. Shot 3 (hands enter frame and pick the product up) and shot 5 (wide desk shot becoming a close-up) end somewhere visually different, so both get two keyframes, with the first and last frames described as held stills, never as actions in progress.
References locked: one presenter still (mid-shot, plain backdrop, wardrobe fixed) drives both presenter shots; one style still fixes the b-roll lighting and palette for all four.
Briefs written per shot with: locked reference, aspect/duration, camera move in plain language, held first frame (and last frame for shots 3 and 5), must-not-change list (presenter identity and wardrobe, product color, lighting direction), and audio expectation.
Audio decision, made in advance: presenter shots request built-in speech with the exact script lines. If the first returned clip carries that speech, it is lip-synced and becomes the deliverable audio — no separate narration track gets synthesized over it. Only if clips return silent do we synthesize narration and lay it under. Captions go on as visual overlays timed to the spoken words, clear of the face.
Verification on arrival, per clip: decodes, duration and aspect match the brief, subject matches the reference, audio state matches the expectation. A mismatch gets logged and returned with the reference attached — never re-rolled on the spot. Scorecard check on the weakest shot (shot 5): brief 2, reference 2, keyframes 2, audio 1 (b-roll, music added later), settings 2, verification 2 = 11/12, clear to generate once you approve the ~12-call budget.
Connects securely to your tools. The creator never sees your data.
About this skill
Generated footage is the most expensive pixel in video production — every clip is a paid, multi-second hosted call, and the classic way to waste the money is to generate first and count later. This skill runs generation as a budgeted discipline. It opens with a cost contract: exact shot count, character count, one aspect ratio per run, and the approximate number of paid calls, approved before anything generates — plus the rule that a produced clip is never re-rolled without fresh authorization. Two production patterns cover most work. The talking-head pattern locks one reference still per presenter and contains the most expensive defect in AI video: if a clip returns with built-in speech, that audio is lip-synced and is the deliverable — synthesizing a separate narration track over it divorces lips from words. Only silent clips get narration laid under. The cinematic b-roll pattern storyboards first and generates per shot against a shared style reference. Keyframe economics decides how many frames a shot actually needs — one when start and end look nearly the same, two when the shot ends somewhere visually different — so you stop paying for end frames that buy nothing. Every shot gets a written brief (template included) covering the locked reference, held first and last frames described as stills, what must not change, and the audio expectation. A readiness scorecard gates each shot at 9+ of 12 before you pay for it, and the worked example runs a six-shot vertical explainer end to end, including the audio decision made from the first returned clip and a wrong-color shot logged and returned instead of silently re-rolled.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.
- It covers the lightweight continuity method (locked reference stills within a run). A character who must stay identical across many scenes, or a real person's photo in the lead, needs the full consistency workflow instead.
- Cost figures are counts of paid calls, not prices — per-call pricing varies by provider and plan, so the budget is confirmed in calls and converted with your provider's rates.
- It does not cover editing existing footage or bounded AI edits to real footage beyond declaring that boundary; those belong to an editing workflow.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected