Portfolio · Director, AI Content Studio — Loblaw Agency
Not a gallery of outputs — a functioning AI content studio in miniature. Production tools you can run right now, built on thirteen models from six labs, with the routing, cost governance and quality gates that make AI production survivable at retail volume.
What the role asks for
The posting is unusually specific about wanting a hands-on producer, not an oversight-only leader. Each mandate below is answered by something in this portfolio you can open and use.
“Ability to produce finished creative content independently across stills, motion, video — not just direct others to do so.”
Everything here is a working tool, not a case-study screenshot. Upload a product photo and it produces finished stills, bilingual promo tiles, GS1 packshots and scored video ads — live, in the browser, in minutes.
“Stand up the AI Content Studio's tools, workflows, standards, intake processes, and quality gates.”
The pipeline is the point: self-filling intake, one brief driving every deliverable, model routing by job and price, and named quality gates. The playbook documents all of it so it scales past one person.
“Evaluate AI tools, platforms, vendor relationships, and build-vs-buy decisions as the production model evolves.”
Thirteen models from six labs are wired in behind two APIs, swappable from one config file, with a written point of view on what's ready, what's emerging and what isn't viable — plus a one-pager on suites vs. aggregators vs. direct APIs.
“Build reusable playbooks, prompt approaches… support responsible AI use, including rights/IP, brand safety, disclosure and approval.”
The generative-AI guidelines this work is held to are written up as decisions rather than clauses, with a pre-flight checklist you tick against an actual asset. Then they are enforced rather than promised: every asset exposes the prompt that made it, reconstructed packshot angles are flagged for label QA, prices are never invented, and the demo states its own limits — because knowing what AI can't do yet is the job.
The tools
Each covers a different slice of what a retail agency actually ships — and they share the same intake, routing, cost and governance layer.
Preset ad recipes any SKU can run through
Deconstructed concepts — aesthetics, beat-by-beat action, overlay spec, sound design — every part of it editable, with the product swapped in from a photo and held still by reference-to-video. Sound is built in layers: effects from the model, spot effects generated one at a time, music scored separately.
One photo → a full multi-format pack
Vision analysis, an AI-written editable brief, then six stills and two videos across formats, languages and seasons — each routed to the right model at the right price.
GS1 planogram angles without the reshoot
Upload the reference angles a SKU already has, generate the missing product-on-white views with GS1 filenames — grounded angles marked, reconstructed ones flagged for label QA.
The anatomy of a reference prompt, as a form
Register, subject, what each reference is FOR, then a timeline that makes you spend the seconds on purpose. Every part says what goes wrong without it, and it catches the mistakes that fail silently — a beat sheet that overruns the render, a reference bound by filename instead of token.
The problem this is built for
A single campaign multiplies across formats, placements, seasons, store banners and — in Canada — two official languages. One idea becomes forty assets before it reaches a shopper. Traditional production can make the one beautifully; it cannot make the forty at that cadence.
That's the gap AI closes, and it changes where craft lives. An adaptation that took a studio day becomes a routed call costing cents, so judgment moves upstream — into the brief, the prompt system and the quality gate. The skill isn't operating the tools. It's deciding what good is, and building a system that produces it repeatedly.
What working with these tools actually teaches you
The role asks for an internal authority on what's ready, what's emerging and what isn't viable. That judgment comes from hitting the limits, not reading about them.
They render SFX, ambience and dialogue convincingly, then approximate music. So the Ad Lab splits the layers: the video model does sound design, a music model scores, and the mix stays a finishing step in the edit.
Most image models still mangle type. That single constraint is why bilingual promo tiles route to the pro tier while format adaptations run four times cheaper on flash.
A generated packshot of a panel no camera captured is a plausible reconstruction, not a record. It gets labelled that way, every time, because a wrong ingredient list is a recall, not a retouch.
Sora 2's API sunset stranded pipelines built on it. Model IDs and prices live in one config file with env overrides, so switching a vendor is an edit — not a rebuild.
Under the hood
| Nano Banana Pro (Gemini 3 Pro Image) | $0.134/image | Promo tiles with live text, bilingual versioning, brand-critical hero shots. |
| Nano Banana (Gemini Flash Image) | $0.039/image | Format adaptations, seasonal variants, bulk versioning at scale. |
| Flux 2 Pro (Black Forest Labs) | $0.05/image | Photorealistic hero stills where the image is the product. |
| Flux Kontext Max (BFL, via fal.ai) | $0.08/image | Packshot challenger: identity-true edits where texture matters more than dense label text. |
| Seedream 4.0 Edit (ByteDance, via fal.ai) | $0.03/image | Packshot challenger: the value benchmark every bake-off should include. |
| Veo 3.1 Fast (Google) | $0.15/second | Drafting, social clips, high-volume video versioning. |
| Veo 3.1 (Google) | $0.75/second | Hero spots and finished broadcast-style deliverables. |
| Kling 3.0 (Kuaishou, via fal.ai) | $0.1/second | Cost-efficient social cutdowns; same product across many shots. |
| Runway Gen-4 (via fal.ai) | $0.15/second | VFX-leaning brand films and shot-heavy storytelling where you expect to iterate on the motion. |
| Seedance 2.5 (ByteDance, via fal.ai) | $0.46/second | Short-form product ads at volume; longer cuts no other model does in one pass. |
| Seedance 2.5 Reference (ByteDance, via fal.ai) | $0.46/second | Product-identity-critical ads: the packaging must stay exactly itself while the camera moves. |
| ElevenLabs Sound Effects v2 (via fal.ai) | $0.002/second | The hero product sound — the crack, the pour, the seal breaking. The one noise the ad is actually selling, which a video model only ever approximates. |
| ElevenLabs Music (via fal.ai) | $0.0133/second | The music bed under an ad. Video models render SFX and ambience well but do not compose music; this layer does. |
Every ID and price sits in one config file with environment overrides — swapping a model is an edit, not a rebuild.
Every page here is open, and demo mode walks the whole pipeline — intake, brief, routing, quality gates — at zero spend. Live rendering is behind a passcode, because each one bills a real account: a 14-second video is roughly $6. Every live run shows its estimated cost and asks before it spends.
If you are reviewing this for the role, I will happily open it up — message me on LinkedIn or email ajwadrauf@gmail.com.
Open the Ad Lab →