Beta feature — Contact your Customer Success Manager for access.
TL;DR
Creative AI automatically labels every image and video creative in your ad accounts by their visual elements (human faces, text overlays, color story, product placement, etc.) and shows you which patterns are actually driving performance. It's prescriptive — at its strongest, it tells you "switch from Pastel/Soft to White/Cream and you can expect +62% CTR lift."
What problem does it solve?
When your creative team builds a brief, the same questions come up every round:
"Which color story performed best last quarter?"
"Do creatives with people in them outperform product-only shots for our category?"
"Are we wasting impressions on visual patterns that don't convert?"
"If we shift from one style to another, how much lift can we expect?"
Today, answering these means manually reviewing hundreds of creatives or relying on memory. Both are unreliable.
Creative AI labels every asset automatically against a taxonomy tailored to your industry, then ties those labels to performance metrics. The result: data-backed input for new briefs and concrete swap recommendations for live campaigns.
How the labeling works
Every creative in your account is analyzed automatically against a label schema customized for your account. The schema covers visual attributes that matter for your category — for a home-goods brand it might include Color Story, Product Category, Table Setting, Occasion Context. For a fashion brand it would look different.
Each creative is tagged with multiple label values at once. For example, a single image might land as:
Color Story: Bold / Vibrant
Food Present: No Food
Table Setting: Product Only
Person Present: No Person
Product Category: Multiple Categories
Labels are color-coded chips so you can read a creative's character at a glance.
Custom taxonomy: how the schema gets set up
The label schema is configured per workspace, not per user. Here's how it works:
Initial setup during onboarding. Customer Success works with your team to design a schema that fits your industry, product range, and the questions you most want answered. If you have an existing internal taxonomy (e.g. how your creative team already classifies assets), share it — we can map it directly.
Schema is applied to your full creative library. Once the schema is locked in, Orphex labels every creative in your connected accounts according to it. Labeling continues automatically as new creatives are pushed to your ad platforms.
Changes go through Customer Success. If you want to add a new label dimension, rename values, merge or split categories, or remove labels that aren't useful — email your CSM with the change request. We update the schema config and trigger a re-label of affected creatives.
Re-labeling is a background job. When the schema changes, existing creatives are re-analyzed in the background. There's a short delay before the updated labels reflect in the Overview metrics — typically a few hours, longer for large libraries.
Self-serve schema editing isn't available in beta. This is intentional: schema changes ripple through every analysis (Comparative, Performance Outliers, Opportunities), so we want a human review step to prevent breaking historical comparisons. Self-serve taxonomy management is on the roadmap for post-beta.
The four tabs
Creative AI is organized into four top-level tabs:
Tab | Status | Purpose |
Overview | Live | Summary metrics + analytical layers (comparative, performance, opportunities) |
All Creatives | Live | Browse and filter the full creative library |
What if | 🔜 Soon | Simulate the impact of label changes before they happen |
Launch Analysis | 🔜 Soon | Predict how a new creative will perform based on its labels |
Across all tabs, two global toggles apply:
Image / Video toggle — switch between image and video assets
Date range — default last 28 days; custom ranges supported
Overview tab
The Overview is where you go for insights. It has two parts: summary cards at the top, then four analytical sections below.
Summary cards
Four cards across the top give you the headline state of your account:
Card | What it shows |
AI Labeled | Total assets analyzed (e.g. 63 images analyzed) |
Most Labeled | The label values that appear most often, with frequency bar (e.g. Person Present · No Person · 56 / 89%) |
Top Performing Label | The single label value driving the highest lift (e.g. Color Story · White / Cream drives +17.39% higher CTR) |
Lower Performing Label | The single label value with the largest drag (e.g. Feature Claim · Capacity / Size drives 36.69% lower CTR) |
These cards are your daily check-in: in 5 seconds, you know what's working and what's hurting.
Comparative Analysis
"Compare performance with vs without a label."
Pick any label-value combination (e.g. Color Story = Dark / Moody) and see two side-by-side cards: creatives with that label vs creatives without it.
Below the cards, a full metric table shows the difference for every metric — Cost, CPM, Impressions, CTR, Clicks, CPC, Leads, CPL — with the difference column highlighted (green = better, red = worse).
Why this matters: Sometimes a label looks like a winner on CTR but is a loser on CPC. The metric grid surfaces those trade-offs before you commit budget.
Filters available:
Platform (All Platforms / Google Ads / and others depending on your connections)
Campaign objective (All Objectives / Purchase / etc.)
Label, Value, Metric, Campaign Type pickers
Label Performance
"Performance across all label values."
For a chosen label (e.g. Color Story), see every value (Bold/Vibrant, Dark/Moody, Earth/Natural, Pastel/Soft, White/Cream) ranked side-by-side with their full metric breakdown.
This is the raw data view — useful when you want to see all options at once rather than just the winner.
Filters:
Platform,
Objective sub-tabs (All Objectives / App Engagement / App Install / Purchase),
Label,
Metric (multi-select up to 5),
Campaign Type.
Performance Outliers
"Best and worst performing label values vs. average."
Two columns:
Top Performing Labels — values that beat the account average on your chosen metric, with the % delta vs average
Lower Performing Labels — values that underperform the average
Each row shows: label, value, the metric value, and how it compares to average — plus a count of creatives carrying that value.
Reading example:
Top: Color Story · White / Cream · 4.06% CTR · +17.39% vs average · 8 creatives Lower: Feature Claim · Capacity / Size · 2.19% CTR · -36.69% vs average · 5 creatives
Filters:
Platform tabs (All / Google / Meta / TikTok),
Objective sub-tabs,
Metric.
Opportunities Analysis (the prescriptive layer)
This is the most actionable section. The table answers: "If you swap label value X for value Y, what lift can you expect?"
Column | Example |
Label | Color Story |
Current | Earth / Natural (8 creatives) |
Switch To | White / Cream |
Current CTR | 2.4% |
Est. CTR | 4.06% |
Potential Lift | +68.9% |
You're not just told what's working — you're told what to change and how much it might be worth. Each opportunity is grounded in your own data; the "Est. CTR" is the historical performance of the target label value in similar campaigns.
How to act on it: Take the highest-lift opportunity, click into the affected creatives (each row links to the underlying assets), and use those as the basis for your next creative brief.
All Creatives tab
The All Creatives tab is your library — every labeled creative, browsable and filterable.
Card view (default)
Each card shows:
Creative thumbnail (image or video frame)
Campaign name with platform icon
Status badge (Active / Paused)
Labels section — top labels with their chips, plus a "View N More Labels" expander for the rest
Objective (App Engagement, App Install, Purchase, etc.)
First Seen date
Best for visual exploration — when you're scanning to spot patterns rather than comparing numbers.
Table view
Same data, columnar layout. Columns: Image, Campaign Name, Objective, First Seen, Active Days, Image Name, then one column per label dimension (Color Story, Food Present, Feature Claim, Table Setting, Person Present, Occasion Context, Product Category).
Best for sorting, scanning many rows at once, or comparing across label dimensions.
Search and filter
Search by campaign name
Side filter panel (icon top-right) for label-based filtering
Image / Video toggle (top of page)
Side panel (creative detail)
Click any creative to open the slide-in panel:
Large preview (image or video)
Campaign name + status badge
Full label list — every label dimension with its assigned value
Show Less / Show More to collapse or expand the label list
Metadata block: Objective, First Seen, Image Name (file)
Performance block: Active Days, Cost, Clicks, Impressions
Use it for fast triage: scan the labels to understand the creative's character, scan the metrics to understand its impact, decide whether to keep, kill, or replicate.
Coming soon: What if & Launch Analysis
Two tabs are reserved but not yet live:
What if — Simulate label changes before you produce the creative. "What if I swap Bold/Vibrant for White/Cream on this campaign — how does my forecast shift?"
Launch Analysis — Upload a new creative or describe its labels; get a forecast of likely performance based on similar past creatives.
Both are tracked in the roadmap; release timing depends on beta feedback on the live tabs first.
FAQ
Q: Where do labels come from? Each creative is analyzed automatically against your workspace's label schema. The schema is industry-specific — a fashion brand's labels look different from a home-goods brand's.
Q: Can I edit a wrong label? Manual override isn't available in the beta UI. If you spot mislabeled creatives, report them via Intercom — manual override is on the roadmap.
Q: Can I add custom labels to the taxonomy? The schema is locked at the workspace level and managed by Customer Success. To add or change labels, email your CSM with the request — we'll update the schema and re-label affected creatives. See "Custom taxonomy" above for the full process.
Q: How often are labels refreshed? New creatives are labeled on an ongoing basis. Existing creatives are re-labeled when the schema changes (background job, typically a few hours).
Q: What's the difference between Performance Outliers and Opportunities Analysis? Performance Outliers tells you what's already winning or losing. Opportunities Analysis tells you what to do about it — which swaps to make and how much lift to expect.
Q: Why is the Date filter set to "Last 28 days" by default? Most performance signals stabilize over a 4-week window. Shorter windows are noisy; longer windows blend in stale data. You can change it.
Q: I see the Image / Video toggle but no videos appear when I switch — why? Video labeling is supported but only for accounts with video creatives in the date range. If you only run image creatives, the Video tab will be empty.
Q: My creative has labels I don't recognize — what do they mean? Hover any label chip in the side panel for the description. If a label's purpose isn't obvious, send the screenshot to Customer Success — taxonomy descriptions are part of what we refine during beta.
Q: Which platforms are supported? The Overview analyses run across All Platforms by default and segment by Google Ads, Meta Ads, TikTok Ads tabs. Coverage depends on which platforms you've connected to Orphex.
Beta limitations
No manual label override in UI (next release)
Custom taxonomy changes go through Customer Success, not self-serve
What if and Launch Analysis tabs are reserved but not yet live
Schema updates re-label existing creatives in the background — short delay before they reflect in the Overview metrics
The estimated lift in Opportunities Analysis is historical-pattern-based; it's a forecast, not a guarantee
Feedback
Your feedback during beta drives what we ship next. Most useful areas to comment on:
Taxonomy fit — does your schema cover the visual dimensions that actually matter for your creative briefs?
Label accuracy — mislabeled creatives, or labels that consistently miss nuance
Opportunities Analysis usefulness — are the swap recommendations specific and concrete enough to act on, or too generic?
Comparative Analysis depth — are the metrics shown the right ones, or should we surface different ones (e.g. ROAS, CAC) more prominently?
What if / Launch Analysis priority — which of the two soon-tabs would be more valuable for your workflow?
Reach out via Intercom or directly to your account manager.