AI Camera

AI Camera restyles your webcam live with generative AI, triggered by what your viewers do. A tip, a cheer, a gift, or a chat command can turn your feed into anime, neon, or cinematic — or drop a glowing dragon over the scene — for a set number of seconds, then snap back to normal.

How it works

You set up rules (we also call them triggers). Each rule says: when this platform event happens, run this prompt for this many seconds. When a matching event arrives and your capture window is open, the prompt runs through the AI model and your feed is restyled for the duration, then returns to your raw webcam.

Rules are the entire configuration — there's no global on/off switch. AI Camera fires only when an event matches a rule you've added. Rules work the same way as the triggers on alerts and overlays, so if you've used those, this will feel familiar. See Elements & overlays for the trigger concept.

Before you start

AI Camera needs two things set up first: a connected bank for payouts (so paid triggers can pay out and refund) and at least one streaming platform connected. If either is missing, the settings page shows a “complete setup” screen instead. Connect platforms in Platform Connections — there's a walkthrough in Connecting platforms.

Landmark tips are always available as a trigger source (they run through Landmark directly). Twitch, YouTube, Kick, Rumble, and TikTok events become selectable once you connect each platform.

Setting it up

Open AI Camera settings. (The old /dashboard/camera link redirects here.)
Add a rule. Pick a platform and event, set a threshold (minimum tip, bits, gift, or a chat command), choose a mode and a prompt, and set how long it should run.
Turn on moderation buckets and add any custom blocked words you want.
Open the capture window and window-capture it in OBS. Your rules fire while that window is open.

Rules in detail

Each rule is built from a few parts. You only see the fields that apply to the platform and event you pick.

Event & threshold: Choose what fires the rule — a Landmark tip, a Twitch cheer or sub, a YouTube super chat, a Kick or TikTok gift, a chat command, and so on. Depending on the event you can set a minimum amount, a minimum count (bits, months, raiders), a keyword/command, or pick a specific gift.
Mode: How the AI changes your feed — see Modes below.
Prompt: Set prompt uses a fixed prompt you write (e.g. “Gaming tiger, extreme lightning, realistic”). User prompt uses the viewer's own message — whatever they type after a chat command, or the message attached to a tip or super chat. User prompt is only available for events that actually carry viewer text.
Duration: How long the restyle runs, from 15 to 600 seconds. (We floor at 15s because the model needs a moment to lock onto a new style.)
Cooldowns: Optional rate limits for events without a money throttle (chat commands, tips, free actions): a global cooldown and a per-viewer cooldown, both in seconds.

Modes

Every rule runs in one of three modes, which decide how the AI interprets your prompt:

Style: Applies a visual style to the whole scene — anime, neon, cinematic, and so on.
Additions: Adds objects or effects on top of the scene — a floating crown, drifting particles, a creature.
Max: Reads the prompt and decides automatically whether it's a style, an addition, or both — splitting a combined prompt as needed. Use this when you want one rule to handle whatever a viewer throws at it.

Pricing & billing

AI Camera is billed by the second. Every second a prompt runs costs $0.03 of Landmark usage, charged to your Landmark wallet and settled with the rest of your usage on the 20th of the month. A 60-second restyle costs 60 × $0.03 = $1.80.

Because most triggers carry a payout (a tip, a gift, bits), the settings page shows a profit/loss chip on each rule: the streamer-net payout minus the usage cost. A green chip means the trigger more than covers the AI cost; a red one means you're absorbing the difference. Free events (raids, plain chat commands) have no payout, so you absorb the whole cost — handy for hype moments, but worth a cooldown.

Tip pricing modes

For Landmark-tip rules you can choose how a tip maps to time:

Fixed: Any tip at or above your minimum fires the prompt for the duration you set.
Pay-per-second: Bigger tips buy more time. You set a price per second (in cents) and a maximum cap, and the tip is converted to seconds — capped at the duration limit.

Payout figures are net of platform cuts (and the 5% Landmark fee on tips). On real tip charges, Stripe card-processing fees also come out of your share — see Where your tip goes for the full breakdown.

Content filters & moderation

User prompts can be anything a viewer types, so AI Camera runs every prompt through a word filter before it queues. A prompt that matches any blocked word is rejected and the viewer is fully refunded.

The Moderation panel gives you canned buckets plus your own words:

Slurs — racial, ethnic, homophobic, transphobic, ableist. On by default.
Sexual — nudity, sex acts, fetish terms. On by default.
Swears — generic profanity and crude insults. Off by default.
Violence — gore, killing, self-harm. Off by default.

Add your own blocked words in the same panel. You can reveal the exact words in any enabled bucket with Show words if you want to see what's covered.

The capture window

The capture window is the surface you put on stream — a clean, full-window webcam feed that the AI restyles in place when a prompt is active. Window-capture it in OBS like any other source. While the window is open it sends a heartbeat, which flips the AI Camera is ON indicator in settings; close it and rules stop firing.

The capture window is deliberately chrome-free — it's an OBS source, so control lives in the dashboard and the control dock, not on the feed. The only thing it draws is a draggable, resizable overlay that shows {Username} - {prompt} | {time} for the active prompt. Its position and size persist between sessions, the edit handles appear only on hover (so OBS captures cleanly), and while editing you can nudge with the arrow keys (⇧ for 10px), press R to reset, and read the live x,y · w×h. Press S to skip the active prompt (50% refund of the remaining seconds).

Output labels

Two settings control what the feed draws on top of the (restyled) video:

AI content disclosure — a small “AI-generated” tag shown while styling is live. Off by default; turn it on if your platform or policy wants AI-altered video labelled.
Landmark watermark — the Landmark wordmark. On by default.

Control from the dock + testing

The control dock (an OBS custom browser dock or your phone) can skip the active prompt, watch the queue, and test AI Camera without leaving your scene — add the AI Camera tab and a Test AI Camera deck button. The same test lives in settings. A test fires a real ~15-second styling prompt against your open capture window, so it runs live inference and bills you — there's no free test (you're warned first).

Spending cap

In Output & control you can set a monthly cap on your Landmark Usage so AI Camera can't run up a surprise bill. When this cycle's usage reaches the cap, your own test fires are blocked until you raise or clear it.

On-stream status element

Want to show what the camera is doing? Add the AI Camera status element to an overlay (Elements → add element) — it displays the live prompt, who triggered it, time left, and the queue depth.

For step-by-step OBS browser-source and docking instructions, see OBS setup.

Set up AI Camera →