The client board signed off Monday at 11:40. One hero. Product front and center, talent locked, palette approved, tagline set. Art direction closed the deck. By Thursday the media plan still needs a 1:1 feed, a 4:5 carousel set, 9:16 Stories and Reels frames, and a short cut around fifteen seconds that reads as the same brand — not a cousin shot regenerated under deadline heat.

That gap is where AI ad creative stops feeling fast and starts feeling expensive. The hero was never the hard part. Shipping one brand across six formats without twelve new prompts is.

Small and mid creative agencies live this week on repeat. Ads and social packs. Campaign variants. Volume, deadlines, and margin pressure in the same Slack channel. Midjourney or Flux already produced the hero. CapCut is open for the cut. Drive has a folder named Client_Campaign_Final_v3.

None of that means Thursday is ready. Meta still wants feed and carousel. Stories and Reels want vertical. TikTok and display may hitch a ride on the same visual system. The approved Monday frame is a single ratio. Every other ratio in the plan is still an unpaid prompt fight if the team treats each crop as a new brief.

The art director does not need another generator seat. They need a pack that keeps the approved face, product, and universe intact while the canvas changes.

The stack is honest about what it is good at. Midjourney and Flux generate. CapCut edits. Drive stores. Ads Manager runs.

What it does not do by default is remember that Monday’s hero is the lock for Thursday’s Stories frame. Change the aspect ratio in a fresh Discord thread and you are back to describing the talent, the wardrobe, the set, the grade. Paste a slightly different prompt into Flux and the product logo drifts. Crop hard in CapCut without locked masters and the composition that won the board collapses into dead space or a chopped shoulder.

Multi-format ad creative AI work fails here for a boring reason: the tools never shared a brand kit. Each hop reopens identity. That is how AI campaign variants agencies ship start to bleed — same campaign name, different face across ratios, client notes that sound like “close but not her.

Industrializing AI advertising creative packs is not about regenerating faster. It is about refusing to reopen the hero once it is approved.

Before anyone derives formats, lock the kit.

Characters: the talent or mascot the board approved — face, wardrobe rules, age read, expression range. Universes: the set, lighting logic, props, and grade that make the campaign feel like one world. Palette and type rules: the colors and wordmark treatments that survive a square and a vertical crop.

This is the creative agency AI workflow most studios skip because Monday’s win feels complete. It is not. Without a lock, Thursday’s vertical is a new casting call with a deadline.

Persistent characters and universes are the boring foundation. Once they exist, every format is a derivation, not a re-brief. Director-style pack orchestration sits on top of that lock: one identity, many canvases, same campaign spine.

Do not rebuild the whole Idea → Create → Produce → Publish → Measure → Improve loop in this pass. Touch it lightly: Create owns the lock; Produce owns the pack. Measure can wait until the set is live. Continuity starts with the kit, not with another Ads Manager screenshot.

With the kit locked, derive — do not re-prompt from scratch.

1:1 feed: rebuild composition for the square safe zone. Keep product and talent hierarchy; move secondary type if the crop demands it. 4:5 carousel: same lock, card rhythm that still reads when someone swipes. 9:16 Stories and Reels: vertical priority — face and product in the mobile thumb zone, not a panicked center crop of the Monday hero. Short cut (~15s): motion and edit that inherit the still pack’s face, wardrobe, and world so the reel does not introduce a stranger.

Multi-format production from one lock is the production rule. Image gen and edit stay inside the same continuity-aware workflow instead of exporting orphans into CapCut and hoping the grade matches. Optional talking avatars belong only when the ad needs an on-brand face speaking — still from the same character lock, not a new cast.

If a ratio needs a fresh generation pass, condition it on the lock. The prompt is “extend this approved kit into 9:16,” not “make a cool vertical ad for Brand X again.”

Thursday delivery fails in Drive as often as it fails in generation. A folder full of untitled exports is not a review. A client review set is one package: the Monday hero for context, the feed, the carousel cards in order, the Stories/Reels frames, the short cut, and a one-line note on what stayed locked.

Art directors review packs. Clients approve packs. Orphans invite piecemeal notes that reopen Midjourney for a single crop and break the rest of the set.

Director-style packaging is the difference between “here are files” and “here is the campaign in every format we sold.” Soft continuity with how agencies close revision loops (https://havincy.com/blog/ai-tools-for-agencies-revision-loop) and how freelancers productize a client kit (https://havincy.com/blog/ai-tools-for-freelancers-client-kit) points the same way: one system of record per brand, not a Discord memory.

Specialist tools still have a job. They sit after the still pack exists — not instead of it.

Kling and Runway extend motion from locked frames or locked characters when the brief needs video beyond a simple CapCut assemble. ElevenLabs covers voice when the cut needs VO that matches brand tone. None of them should become the place where you reinvent the Monday hero because the still pack never got locked.

Sequence matters. Lock kit → derive still pack across ratios → then motion and voice. Reverse that order and you get beautiful clips that refuse to sit next to the feed square the client already loved.

Count the Thursday hours honestly.

Path A: reopen Midjourney or Flux per ratio, re-explain the talent, re-grade in CapCut, re-export, rebuild the client PDF, absorb brand-bleed notes, re-export again. Path B: one locked kit, one derive pass across 1:1 / 4:5 / 9:16 / short cut, one packaged review set.

Path A burns margin on work the Monday approval already paid for. Path B spends hours once on continuity and spends the rest of the week on media, copy, and client conversation. For a small or mid agency shipping ads and social variants under deadline pressure, that difference is the line between AI seats that feel productive and AI seats that feel like unpaid overtime with prettier first drafts.

Lock characters, universe, and palette before volume. Derive feed, carousel, Stories/Reels, and the short cut from that lock. Package one review set so notes hit the campaign, not orphan files. Keep Midjourney, Flux, CapCut, Kling, Runway, and ElevenLabs as specialists inside that continuity — not as twelve separate briefs that happen to share a campaign name.

Yes, often. CapCut (or Premiere, or any editor) still assembles, captions, and times the cut. What changes is the input: locked masters and derived frames instead of a Drive dump of cousins. Edit stays. Recasting the brand inside the edit dies. Image plus edit plus video in one continuity-aware workflow is the point — CapCut remains a station on the line, not the place where identity gets reinvented under Thursday heat.

One approved hero is a win. Six formats that still read as that hero is the product. Stop treating every ratio like a new Midjourney morning. Lock the brand kit once, derive the multi-format pack, package one client review set — and let motion tools earn their keep after the still pack already holds.