×
H2Innovation
Turning production pressure into creative opportunity.
Innovation only mattersif it solves a real problem.
The opportunity is not simply to “use more AI”. It is to identify where new tools, workflows and production models can make Meta’s global content operation materially better.
How the session is structured.
The H2 challenge
Innovation should remove friction,not add another layer of complexity.
Pragmatic rather than dazzled by technology.
H1 moved AI from theory into live production.
What those pilots taught us.
Five H2 objectives
Not tools. Not initiatives. Objectives, because each one solves a business problem.
Build once.
Scale intelligently.
Production logic gets rebuilt from scratch every campaign. Build it once, and every market after that is a rules change.
The build carries over, so a new market starts from work already done rather than from scratch. Rollout moves closer to a configuration than a new production.
Intelligent campaign automation - plan, create, validate and activate. Publishes straight to Meta, YouTube, DV360, TikTok, Amazon and DOOH. Used by Unilever, Ubisoft, Paramount, KLM and Philips.Walkthrough to come
ComfyUIA fully reusable GenAI workflow that swaps faces, clothing and background while holding the composition of the image.In build
3 placements × 3 markets
These are the real Citrine masters and their India and Mexico localisations, reframed for each social placement. Nine finished assets from one build.
Keep the human.
Reduce the shoot footprint.
Shoot real talent once, in a controlled studio. Then build every scene, market and format from that single capture.
Real-talent credibility in every market, without paying for a location shoot in every market.
Shoot once. Build the rest around it.
The capture is real and the creative direction is human. What changes per market is everything around the performance, not the performance itself. There are two ways to put a world behind that performance, and only one of them keeps it editable.
The background stays a variable rather than a decision made on shoot day. The shoot happens once, and every market after it is an assembly we build and run.
Generate movement,
not poses.
Phase 2 showed that natural human behaviour is difficult to manufacture through static prompting alone. So we stop prompting for a pose and direct a performance instead - then take the frame we want out of the motion.
Social creative that reads photographed rather than generated - from creative you have already approved.
Movement, not a pose
Because it starts from creative Meta has already approved, there is no new brief and no new risk - only the difference between a still that reads posed and a frame that reads photographed. This is promise 02 from Phase 2, in production.
One performance.
Many markets.
Move the conversation beyond “AI dubbing”. Start with a strong human performance, keep what makes it good, and scale it intelligently across markets.
BeebleAIKeeps the original performance and adapts the character for the market - same performance, different person. Runs entirely on local hardware; source files are never uploaded to the cloud.Tested
Sync LabsThe voice side: dubbing and lip-sync that carries one performance into every language.Demo video to come
Automatic dubbing, already embedded in our process. The baseline we are building past.In our processOne casting and one performance serve every market. Localisation stops being a reshoot.
Hero human-led · creator and social AI-supported · high-volume automation-first.
BeebleAI
Freedman internal test · not client work
The performance stays. The person can change.
One captured performance can carry into a market that needs a different face, without recasting, rebooking or reshooting. The timing, the delivery and the intent are all preserved - only the talent changes.
The voice works the same way. Sync Labs re-records the line in the target language and re-times the mouth to match it, so a dubbed market still reads as performance rather than as dubbing.
A market that needs a different face or a different language no longer needs a different shoot. Casting and voice stop being production costs and become settings.
Scale quality control
alongside production.
If production volume grows, quality assurance has to grow with it. An automated pre-check takes the repetitive pass, so human Brand Guardians spend their time on judgement.
Claude pluginOur own internal test automating transcreation, brand guardianship and quality control in one pass. Built in-house; not yet run on a live campaign.Internal test
Clearance, before legal ever sees it. Adclear reads the brand guidelines and the market rules, then checks every asset against them - advertising restrictions differ market by market, alcohol being the obvious one. It pre-approves at volume and flags only what needs a human decision.In pilot with Meta
Eye-tracking heatmaps in seconds, plus one score covering attention, cognitive demand, memorability and focus - and the specific fix. Built on eye-tracking and brain-scanning data from 300,000 participants over 20 years. Already used by Mars, Chanel, L’Oréal, TikTok and Meta.To licenceVolume can grow without the QA team growing with it - and without brand risk growing either.
Product accuracy - pass
Headline 43 / 40 characters
Logo clear space - pass
Transcreation - term not approved
Format 4:5 - compliant
Machines take the repetitive pass. People take the judgement.
Every asset is checked against the brand and language rules before a human opens it. The Brand Guardian then reviews only what was flagged, rather than everything that was made.
on this asset
Neurons AI predicts where the eye lands before anything launches, scores the creative, and names what is costing you attention - weak hierarchy, poor logo contrast, reduced readability. It is A/B testing that happens before the spend, rather than after it.
The compliance pass is our own internal Claude plugin test, not yet run on a live campaign. The attention map shown here is an illustration of what Neurons AI returns, applied to our own master so you can read it against work you know. We would like to pilot both properly with you.
Ask the hologram.
A live holographic avatar with a brand personality, a voice and a product knowledge base. Someone walks up and asks it a question, in their own language, and it answers from copy that has already been signed off.
Partner found. Costed. The right time to scope.
Ray-Ban Meta is a product you talk to. This puts that same conversation in the room, so someone understands the glasses by having the interaction rather than reading about it.
The ask: one activation, one site, two languages.
“These are Ray-Ban Meta. Ask me anything about the frames, the fit or the features.”
From innovation day to innovation pipeline.
Which real programme problems create the most friction?
Which are worth solving through technology or workflow?
Run small, clearly bounded tests.
Cost. Time. Quality. Approval. Scale. Creative outcome.
If it works, integrate it into live programme delivery.
Pass the learning back to the wider Meta team.
What we commit to.
Only ideas we believe solve a meaningful production or creative challenge.
Problem, hypothesis and success measures defined before we begin.
If something does not work, we will tell you. The learning still has value.
Best technology for the problem, not everything anchored to one platform.
Creative, brand and cultural expertise remain central.
Successful pilots become improved workflows, not isolated demos.
Programme managers should not have to spot every opportunity themselves.
Where should we place the next bet?
Which of these pressures is most acute for you in H2?
Which upcoming brief gives us the best opportunity to test something differently?
Where are you currently seeing production friction that we may not see?
×Better technology is not the ambition.Better global creative production is.
Our role is to help you test what is possible, and turn what works into everyday advantage.
Why us?
Because innovation at global scale needs more than a tool.
×
Questions?
Which brief, and which pressure, do we start with?
×