From shopper insight to an AI-assisted creative system

I connected shopper questions, content planning and AI-assisted production into a creative system for 13 Fifty’s mesh caps and beanies. My role was to decide what the work needed to communicate, where AI could help, and when an image needed a different approach.

13 Fifty Apparel, via agency partnersCreative leadCreative strategy / Content planning / AI workflow / Art direction2026

Full size

Current beanie lettering direction · selected October 20, 2026. Product photography and detail-led layout; selection of a design direction is separate from publication approval.

A running process

Research → exploration → direction.

A record of the decisions behind the work. Each step connects the source material, the role of AI and the judgment that moved the project forward.

Updated 21 October 2026 · Work in progress · Scroll position is not project completion

10 SEP 2026Research documented
Separate search traffic from shopper intent
The creative brief combined review themes, keyword and category research, and a listing audit. “Police hat” searches also surfaced costume products; that overlap did not make them the right positioning reference for duty-oriented apparel. The strategy was to make product differences visible and answer the questions that mattered to the intended shopper.
Trace the decision
Source input
Review themes, category and keyword research, current listings and competitor listings.
AI-assisted step
The source brief is a research synthesis. Its precise AI contribution is not recorded, so I do not attribute the underlying research to a specific model.
My decision
Prioritize evidence shoppers can see: lettering, construction, fit, mesh and color options. Avoid turning unverified durability or comfort promises into persuasive copy.
Output / evidence
A content plan organized around product questions rather than a sequence of interchangeable lifestyle scenes.
Milestone 1 of 5.
30 SEP 2026Creative iteration documented
Give each image a job
The mesh-cap sequence moved toward feature-first headlines, detail insets and explicit colorway labels. AI-generated wearers made an otherwise missing use context possible, but the scene still needed art direction. Generated shirt graphics could read as additional products for sale, so I removed that distraction and kept the cap as the subject.
Trace the decision
Source input
Brand product photography, a gap in wearer imagery, and the planned listing sequence.
AI-assisted step
Gemini and ChatGPT image tools supported generated wearers, settings and edits; the private case records Photoshop and Lightroom for retouching and Figma for layout. It does not tie every image to a precise model version.
My decision
Keep the product story legible. Remove misleading generated clothing graphics, make feature hierarchy explicit, and judge the sequence at a 350-pixel phone width.
Output / evidence
A feature-led image sequence and a clearer division between generation, retouching and layout.
Milestone 2 of 5.
01 OCT 2026Workflow tested · mixed results
Design the AI workflow, not just the prompt
Color variants needed consistency without accumulating errors. The local Gemini workflow paired the same original scene with one exact product reference per job. Each output started from those inputs rather than the previous generated result. That made variants comparable and failures easier to trace.
Exploration returns to a decisionGenerated recolor → lettering mismatch → revised constraint or direct retouch → human review
Trace the decision
Source input
Original scene plus the exact product photograph for the requested colorway.
AI-assisted step
Independent reference-based editing jobs shared a prompt template. The revised instruction made the reference authoritative for lettering color, weight and spacing, and treated an unchanged result as a failure.
My decision
Use a pilot gate before expanding variants. Review lettering and construction myself; automated acceptance is not design approval. When a generated edit would not respect the product, change the method rather than accept a plausible-looking mismatch.
Output / evidence
A repeatable job structure with input hashes, prompt/configuration records, output sidecars and explicit review state. This documents traceability, not a measured productivity gain.
Milestone 3 of 5.
20 OCT 2026Direction selected
Extend the system to beanies
The work now spans more than the original mesh cap. The current beanie system separates product views, lettering, fabric and use context into distinct image roles. Shared Figma components carry the layout across colorways while product photography supplies the specific detail. The lettering card puts a close-up beside the product so its message has visible support.
Current image 3 · direction selected October 20. Clean artwork exported from Figma; internal review labels excluded.
Trace the decision
Source input
Current beanie product photographs and the established listing/A+ layout system.
AI-assisted step
AI-assisted production remains one part of the workflow. This lettering card uses product photography and layout; it is not evidence that every asset was generated.
My decision
Choose the lettering-and-product direction for image 3, and keep alternate concepts distinct from the selected sequence.
Output / evidence
A selected lettering direction and reusable per-color layouts. Full-set approval and launch are separate milestones.
Milestone 4 of 5.
21 OCT 2026Current · in progress
Make the material the message
The current fabric direction gives the photograph the full frame and places copy over it, bringing the material forward. It is part of the same system as the lettering card, not an unrelated visual experiment. The current Working page keeps per-color listing sets, shared components and phone-size checks together so the next decision can be made in context.
Current Green & Black fabric direction · October 21. In-progress design candidate; not client approval or evidence of launch.
Trace the decision
Source input
Current fabric photography, the shared Fabric variant and neighboring lettering cards.
AI-assisted step
The system supports AI-assisted exploration, but the current fabric card’s source is product photography. I do not present this layout as an AI-generated product or an approved final campaign.
My decision
Evaluate the edge-to-edge material treatment, contrast and readability as a design candidate. Keep design selection, factual checks and publication clearance separate.
Output / evidence
A current fabric-led candidate for review. The copy’s material, fit and care claims still need final factual clearance before public release.
Milestone 5 of 5.

Next gates · not completed

Approval, launch and outcomes remain separate.

The next updates need evidence: an approved revision, a confirmed handoff, a matching live listing, or measured results with a defined period and source. A new export or a successful AI job alone does not establish any of those states. No sales, conversion or time-saving result is claimed here.

From shopper insight to an AI-assisted creative system

13 Fifty Apparel, via agency partnersCreative leadCreative strategy / Content planning / AI workflow / Art direction2026

Research → exploration → direction.

Separate search traffic from shopper intent

Give each image a job

Design the AI workflow, not just the prompt

Extend the system to beanies

Make the material the message

Approval, launch and outcomes remain separate.

Research → exploration → direction.

Separate search traffic from shopper intent

Give each image a job

Design the AI workflow, not just the prompt

Extend the system to beanies

Make the material the message

Approval, launch and outcomes remain separate.