A content engine that produces festival campaigns two weeks ahead
Spice brand
For a spice brand, Prospicience built a festival content engine: four reusable master templates, a named AI agent for each role, a festival calendar set for the year, and every batch produced 14 days before the festival. Words are added at the design stage, never generated into an image, and nothing publishes without its QA checklist and a person's sign off.

SpeedSet up and first batch drafted in one afternoon
- reusable master templates
- 4
- lead time before each festival
- 14 days
- text generated into images
- none, by rule
- review before publishing
- QA checklist per batch
- 01
Designed four reusable master templates
- 02
Set up a named agent per role, an orchestrator plus caption...
- 03
Scheduled every batch 14 days before its festival, against a...
- 04
Generated product imagery from the brand's own product...
Text is never generated into an image. Every word is set at the design stage, so prices and brand names on a published ad are never distorted. Nothing publishes until it passes its QA checklist and a person signs off.
The challenge
Festival creative was produced days before each festival, which left no time for review, and every festival started again from a blank page.
Why it mattered
Festivals are the brand's biggest sales moments and they arrive on fixed dates, yet the creative was always the last thing ready. Rushed work gets approved by default, and a wrong price, claim or distorted brand name on a published ad is a commercial problem, not a cosmetic one. Starting from scratch each time also meant paying for the same thinking again and again.
What we built
- Designed four reusable master templates: a festival greeting with packshot, a recipe carousel, a reel storyboard, and a where to buy sales post.
- Set up a named agent per role, an orchestrator plus caption, script, graphics and QA agents, with a production board and QA checklists.
- Scheduled every batch 14 days before its festival, against a calendar set at the start of the year.
- Generated product imagery from the brand's own product photographs as references.
The result
- Each festival starts from a proven structure instead of a blank page.
- Every batch is scheduled two weeks early, which leaves real time for review.
- Every post passes a checklist covering product accuracy, claims, marketplace availability and phone readability, with a person giving final approval.
- No ad carries distorted text, because no text is ever generated into an image.
Built for trust
Text is never generated into an image. Every word is set at the design stage, so prices and brand names on a published ad are never distorted. Nothing publishes until it passes its QA checklist and a person signs off.
AI at work
AI agents draft the captions, reel scripts, storyboards and post layouts for every festival, and make product imagery from the brand's own photographs. Designers set every word, and a person approves each batch before it goes out.
What it produces
- Captions, hooks and hashtags
- Reel scripts and shot lists
- Carousel and post layouts
- Product imagery from the brand's own photos
- A QA checklist per post
How it works, day to day
The festival calendar sets the date. The orchestrator assigns the batch, specialist agents draft captions, scripts and layouts from the master templates, and the QA agent checks every post. The design team sets the words, marketplace listings are verified, and a person gives final approval.
Where it saves time, and gets better with use
- Every batch is ready 14 days before its festival, which leaves real time for review.
- Four master templates carry from festival to festival, so each one starts from a proven structure instead of a blank page.
- A QA agent checks product accuracy, claims, marketplace availability and phone readability before a person signs off.
- Setup and the first batch of drafts took one afternoon.
By the numbers
- 4
- reusable master templates
- 14 days
- lead time before each festival
- none, by rule
- text generated into images
- QA checklist per batch
- review before publishing
- 4 specialists plus an orchestrator
- AI agents in the pipeline
The technology
- A calendar of festivals set at the start of the year
- Four reusable templates instead of starting fresh each time
- Every batch ready two weeks early
- Words are added by a designer, never generated into the picture
- Seedance
- GPT Image
Technical notes
- Reference conditioned image generation from the brand's own product photography rather than text prompts alone
- Typography applied at the design stage over generated backgrounds, never inside generation
- A defined agent roster with a production board and per stage QA checklists
- Calendar driven batching with a fixed lead time per festival
Questions we get asked
- Why not let the AI generate the text on the image?
- Because generated type distorts, and a price, offer or brand name has to be exactly right. Generating the image and setting the words separately removes that whole category of error for very little extra work.
- How do you keep AI campaign creative on brand?
- Reference images from the brand's own product photography, reusable templates, and a written constraints file covering colour, type and what the brand must never look like. Every batch starts from the same anchors, so the tenth festival looks as on brand as the first.
- Why produce two weeks ahead?
- Because it guarantees review actually happens. Two weeks gives the team time to check every product name, claim and marketplace link before anything reaches an audience.
Related systems

Brand rules written so AI tools follow them
Used across our client projects
One rules file per brand that designers and AI tools both follow.
6 areasrules per brand

Media Studio: podcasts, dubbing and video from one brief
Prospicience product for creators and media teams
One brief in, a finished podcast, dub or video out.
10+AI services connected

An agentic film studio: brief to storyboard to shots to finished cut
Prospicience internal studio
A brief becomes a finished short film, each shot made by the best-suited model.
4providers routed
First finished cut in under two hours
