Studio Edition
AI Commercial Production Suite
Brief to approved AI commercial: concepts, storyboard, prompt packs, QC and cutdowns
For freelancers, small studios and in-house brand teams producing commercials and product ads with AI image and video models.
Most AI ads fail in three places: there is no idea behind the pretty shots, the product on screen is subtly wrong, and the 16:9 master can't become a 9:16 six-second ad. This skill makes your agent run the job the way an agency does, with a gate at every stage.
- Brief → strategy → three routes: an intake question bank, an insight and proposition one-pager with tests, and three concept routes built on different creative engines, each with an AI-feasibility note, scored and recommended.
- Script and storyboard: an AV two-column script with VO word budgets, and a storyboard CSV where every shot has a role, a source (AI video, AI from still, real photo, composite, graphics), a model and a ratio plan.
- Prompt packs: key-visual prompts for GPT Image, Nano Banana, Seedream, FLUX and Midjourney, and shot prompts for Seedance 2.0/2.5 and MiniMax H3 with a shared continuity bible.
- Product accuracy: a truth sheet, a hybrid composite plan for labels and end cards, and a QC sheet for shots, claims, rights and AI disclosure.
- Delivery and client: a deliverables matrix (9:16, 4:5, 1:1, 16:9; safe zones; file names) with a script that proposes cutdowns, a client deck outline, a revision log with change-impact rules and a handoff note.
Included: SKILL.md, 7 playbooks, 13 numbered templates, a dependency-free Python script and 2 worked examples. The skill writes plans and prompts; you need your own access to the image and video models. Model output varies, safe zones and platform rules change (verify before trafficking), and claims guidance is not legal advice.
Highlights
- Ten-stage agency pipeline with approval gates, from brief to final handoff
- Three concept routes built on different engines, scored and recommended
- Key-visual and shot-by-shot prompt packs for image models, Seedance and H3
- Product truth sheet, hybrid composite plan and frame-by-frame QC sheet
- Deliverables matrix script: ratios, safe zones, file names, 15/6 s cutdowns
Example
Launching a 250 ml cold brew can in UK supermarkets. Need a 30 s film plus social cuts. We have GPT Image 2.5, Seedance 2.5 and H3. Photos attached.
Stage 3/10 · Concept routes · Mode: Full production Proposition: A proper café coffee, ready the moment your day starts — whatever the clock says. | Route | Engine | Idea | |A · The 3 a.m. Café|Human story + metaphor|A night-shift nurse's break room warms into a café for one sip| |B · Eighteen Hours|Ritual / origin|The 18-hour steep as one macro journey of coffee and copper light| |C · Clockless|World-building montage|A city where every morning starts at a different hour| Scores /35: A 30 · B 24 · C 27. Recommend A: it dramatizes the proposition instead of describing taste; the light shift generates well; the riskiest shot (hands on the can) is a short H3 shot, and the end card is a composite of your real packshot. Fallback: C. Gap: "less bitter" has no proof on the truth sheet, so it isn't used yet.
Write the Seedance 2.5 prompt for S04, the can hero shot, using our approved key visual KV-02 as the first frame.
[S04 · Seedance 2.5 · 16:9 · duration 4 · first frame = KV-02] @Image 1 = first frame (approved key visual KV-02). @Image 2 = the product (shape, label, colors). Night-shift drink commercial: the can on a steel break-room counter at 3 a.m. 0–4s: [MCU, eye level, slow push-in toward the can, ending at the label] Condensation beads form; one drop runs down the side. The can stays perfectly still, label facing camera. Look: 35mm photographic, soft contrast, fine grain. Light: cool fluorescent, faint warm glow from frame left. Sound: room tone, fridge hum. No music. No dialogue. No subtitles. Continuity: the can from Image 2 keeps its shape, colors and label text exactly; it does not move. QC note: take 1 misspelled the label at 0:02.6 as the push tightened. Take 2 ends the push at MCU and passes.
python scripts/deliverables_matrix.py --client Kettlebird --project "3am Cafe" --master 30 --cutdowns 15,6 --placements tiktok,youtube-instream,youtube-bumper --storyboard 06-storyboard-shot-list.csv
{
"deliverable_count": 7,
"deliverables": [
{"deliverable_id": "D02", "placement": "tiktok", "ratio": "9:16", "length_s": 15,
"safe_box_x": "65-950", "safe_box_y": "269-1248", "cutdown_shots": "S01;S04;S05;S08",
"file_name": "KETTLEBIRD_3amCafe_15s_9x16_tiktok_v01.mp4"}, …],
"cutdowns": {
"15": {"total_s": 13.0, "slack_s": 2.0, "warnings": ["2.0 s unused: extend the hero or end-card hold, or add a shot"]},
"6": {"total_s": 6.0, "slack_s": 0.0, "shots": [{"shot_id": "S04", "role": "PRODUCT", "duration_s": 4.0}, {"shot_id": "S08", "role": "ENDCARD", "duration_s": 2.0}]}
}
}What's inside
ai-commercial-production-suite/ ├── agents/ │ └── openai.yaml ├── examples/ │ ├── 01-cold-brew-launch-full-production.md │ └── 02-serum-social-fast-track.md ├── references/ │ ├── brief-and-strategy-playbook.md │ ├── client-presentation-and-revisions.md │ ├── key-visual-image-prompting.md │ ├── platform-deliverables-and-cutdowns.md │ ├── product-accuracy-guardrails.md │ ├── script-and-storyboard-craft.md │ └── video-shot-prompting.md ├── scripts/ │ └── deliverables_matrix.py ├── templates/ │ ├── 01-intake/ │ │ ├── 01-creative-brief.md │ │ └── 02-product-truth-sheet.md │ ├── 02-strategy/ │ │ ├── 03-creative-strategy-one-pager.md │ │ └── 04-concept-routes.md │ ├── 03-script/ │ │ ├── 05-script-av-two-column.md │ │ └── 06-storyboard-shot-list.csv │ ├── 04-prompts/ │ │ ├── 07-key-visual-prompt-pack.md │ │ └── 08-video-shot-prompt-pack.md │ ├── 05-qc/ │ │ └── 09-product-accuracy-qc-sheet.md │ ├── 06-delivery/ │ │ └── 10-deliverables-matrix.csv │ └── 07-client/ │ ├── 11-client-presentation-outline.md │ ├── 12-revision-log.csv │ └── 13-final-delivery-handoff.md ├── LICENSE.txt ├── README.md └── SKILL.md
Install by unzipping into your agent's skills folder. Install guide →