Every category says UNLIKELY or below. The photo is almost certainly fine. It goes to the wall.
FEATURE · SMART MODERATION
Your moderators sleep tonight.
Photos hit your queue with a safety score and a brand-fit verdict attached. The 80% that are obvious go through. Your moderators handle the edges.
- Reclaim hoursper moderated event
- 50,000-photofestivals now feasible
- Human-in-the-loopby default
THE PROBLEM
Manual moderation does not scale past dessert.
A 2,000-guest gala at 11 PM produces a queue 400 photos deep before dessert. A festival with 50,000 photos across a weekend is not even on the table without help. And the photos that need a moderator's attention (the off-brand selfie, the competitor's logo visible in the background, the one image the venue's legal team would not want on a wall) are buried under hundreds of obvious approves.
The pain is not the bad photos. It is the time spent clearing the obvious ones to get to the bad photos. Manual moderation works at 60-80 photos per hour per moderator. A gala over 500 photos needs either a second moderator on rotation or an admin who is not enjoying the evening.
Smart moderation is the answer to "how do I stop reviewing photos of the cake one at a time?" It is not the answer to "how do I remove the human entirely?" The AI handles triage. Your moderator handles judgment.
Three steps. The first two happen without you.
- 01
Upload
Guest scans the QR, uploads a photo. Same flow that has always worked.
- 02
AI scores
Within about a second, Google Cloud Vision returns a safety score across five categories. Brand-fit policies are in private beta on Signature+; when enabled for your account, a brand-fit pass runs your event's natural-language policy against the image.
- 03
Admin decides
Obvious approves go straight to the wall. Obvious rejects get logged with a reason. Anything in between lands in your queue with the AI's reasoning visible. Your moderator approves or overrides in one click.
By default, AI runs in shadow mode for the first week. It scores every photo but doesn't change anything until you've seen the results. Move to auto when you trust the calls.
What auto-approve actually means.
The AI is intentionally conservative. We use Google Cloud Vision's 5-level scale (VERY_UNLIKELY through VERY_LIKELY) across five categories: adult content, violence, racy, medical, spoof. Plain-English translation below.
Anything in between. Your moderator sees the photo with a sidebar showing the AI's scores: green bars for each category, amber on the one that hesitated. One click to approve or override.
Any single category says VERY_LIKELY. The photo is logged in the audit trail with the reason and never reaches the wall.
We picked the thresholds tight on purpose. False-approve is the worst failure mode. Better to send a borderline photo to a human than let an unsafe one onto a sponsor's screen. You can loosen the thresholds per event if your moderators want fewer interruptions.
Write a sentence. The AI enforces it.
The safety floor catches the unsafe. The per-event brand policy catches the off-brand. No keyword lists, no taxonomy, no training data: just the policy you want the AI to read on every upload.
-
WEDDING (STRICT)
No smoking. No alcohol bottles or kegs in frame. No children's faces visible.
What the AI does: Flags photos with cigarettes, beer cans, or recognizable child faces; sends them to your queue with the reason.
-
CORPORATE LAUNCH
Formal business attire only. No competitor logos (Acme, Bigcorp, Initech). No half-eaten food.
What the AI does: Flags casual-wear shots, anything with the named logos in the background, leftover plates.
-
FESTIVAL (RELAXED)
Anything goes within the safety floor. Block only obvious gang signs and political messaging.
What the AI does: Light touch. Only intervenes on the specific exclusions you wrote.
-
BRAND ACTIVATION (AGENCY)
Acme Cola launch. No competitor brands visible (Pepsi, Coke Zero, Mountain Dew). No alcohol. Faces only from the approved talent roster of 12 listed names.
What the AI does: Flags any competitor can or bottle in frame, alcohol, and any face that is not on the roster — sends them to your client-approval queue with the reason.
AGENCY USE CASE
If you run activations for clients, this is the per-client policy editor that scales across your roster. Pre-approve photos by sponsor logo or brand-mark visibility, block every competitor brand by name, and gate faces to an approved talent roster — one policy per client, in their words. Each client gets their own moderation rules and their own approval roles, so the brand team that signs off on the Acme launch never sees the Bigcorp queue. It is the wedge against generic photo-booth tools: a natural-language brand policy per event, not a fixed keyword list.
SEE THE AGENCY TIER →The AI reads the policy out loud to you in the editor. If it interprets "no alcohol" as "no glass containers at all," you see that before the event starts. No surprise rejections at 9 PM.
Built for the events where moderation actually matters.
- GALA / FUNDRAISER
Manage 500-photo events solo.
A 300-guest gala used to need a moderator parked at the back of the room. With smart moderation, the planner glances at the queue every 20 minutes and clears the 6 photos the AI flagged. The other 490 went straight to the wall. The night runs.
FOR GALAS → - FESTIVAL / ACTIVATION
50,000-photo events become possible.
A weekend festival generates 30-50k uploads. Manual triage is not in the cards. Even with a 3-person moderation crew, the queue would never clear. Smart moderation handles the safety floor at upload time; the crew reviews the ~5% the AI is unsure about. The wall keeps moving.
FOR FESTIVALS → - CORPORATE / BRAND ACTIVATION
Brand policy enforced automatically.
Marketing wrote a policy: "no competitor logos, no informal attire, no half-eaten plates." The AI catches the casual-dress photo from the after-party and the keynote shot with a Bigcorp banner in the background. CMO sleeps; brand stays intact.
FOR CORPORATE EVENTS →
PRICING
Smart moderation is included on every paid plan.
The safety floor (NSFW + violence + drug detection) ships on Essential, Signature, Premier, Activation, Agency, and Enterprise at no extra cost. Brand-fit policies (natural-language per-event rules) are in private beta on Signature and above; contact us to enable for your account. 50,000 AI calls per month per tenant by default. Most events use about 1,000.
Smart moderation, answered.
-
Will the AI replace my moderators?
No, and we would not design it to. Smart moderation handles the triage (the obvious safety calls and the policy-violating edges) so your moderator's time goes to the photos that actually need judgment. A human is in the loop on every event by default. You can opt into full auto-approval per event if you trust the calls, but shadow mode is where every account starts. Agency teams running activations for clients can configure per-client policies and per-client moderator approval roles, so each client's brand team only sees their own queue. -
What happens when the AI is wrong?
Two scenarios. (1) The AI auto-rejected a photo that was actually fine. We log every auto-rejection with the AI's reasoning to an audit trail, and a sampling tab shows recent auto-rejects so you can spot-check and override. (2) The AI auto-approved a photo that should not have made the wall. The regular moderation queue still works; pull it down with one click and we keep that signal to improve the policy. Either way, the failure is recoverable and visible. -
Is it private? Who sees my photos?
Photos are sent to Google Cloud Vision (the same Google Cloud you trust with your event's storage). The brand-fit pass (currently in private beta on Signature+) additionally sends the image to Anthropic Claude when enabled for your account. Neither vendor retains your photos for training: Google's Vision API is stateless on the SafeSearch endpoint, and Anthropic's API does not train on customer inputs by default. The DPA on Signature and above covers both. Full vendor list is on our subprocessors page. -
Can I see what the AI is looking at?
Every photo in your moderation queue shows the AI's reasoning as a sidebar: a green / amber / red bar per category with the score, and on Signature+ events with a brand policy, a one-sentence note (e.g. "Flagged for: visible logo, brand 'Acme'"). Auto-rejected photos go to a separate "Rejected by AI" tab with the same reasoning attached. Nothing is hidden. -
What does it cost?
Included on every paid plan. We cap usage at 50,000 AI calls per tenant per month. That covers ~50 mid-sized events or a few large festivals. If you somehow hit the cap, we email you before billing anything; we don't run silent overages. -
How does it learn our brand?
It does not learn in the model-training sense. We don't fine-tune anything on your photos. The per-event brand policy is just natural language you write in event settings ("formal black-tie only, no children's faces, no logos of Acme or Bigcorp"). The AI re-reads that policy every time it scores a photo. To "improve" the policy, you edit the text. No training data ever leaves your event.
SEE IT IN YOUR DASHBOARD
Try it on a low-stakes event first.
Free trial activates the safety floor in shadow mode. You see what the AI would have done, without any auto-action. Move to auto when you trust it.