Omni Talking-Actor Model

Omni is the talking-actor model in Polaris — the engine behind AI spokesperson ads. It renders 8-second clips at a flat 120 credits, in 720p or 1080p and 9:16 or 16:9, lip-syncing exactly the script you provide and accepting up to 7 pooled reference images to ground the shot.

What Omni is best at

Script-driven delivery. Where Kling and Seedance excel at product motion and Veo at cinematic scenes, Omni exists for the moment a person looks into the camera and says your line. Testimonial-style openers, problem-solution pitches, direct-response hooks — the formats that dominate UGC-style ad accounts are exactly what it renders. You write the words (or have Claude draft them), and the actor's mouth matches them.

The pricing is refreshingly simple: 120 credits flat, whichever resolution or ratio you choose. That makes budgeting a batch of spokesperson variations trivial — see pricing for how many fit in a monthly pool.

Specs at a glance

• Duration: 8 seconds
• Cost: flat 120 credits
• Resolutions: 720p, 1080p
• Ratios: 9:16 or 16:9
• Delivery: lip-syncs the script you provide
• References: up to 7 pooled reference images

Trade-offs to know

Eight seconds fits roughly one strong line, not a monologue — longer pitches mean stitching multiple clips or pairing Omni with footage from the other Polaris models. Polaris builds ads around your product with generated actors: there is no stock-avatar library to browse, and it will not clone named real people, so you cannot ask for a specific celebrity's face. There is also no 4K tier and no 1:1 ratio. And a lip-synced delivery is still an ad hypothesis — test scripts and actors against each other rather than assuming the first render is the winner.

How to use it

1. Connect the Polaris MCP once, then ask Claude in plain chat — paste your script and attach product or reference images.
2. When the settings question appears as tappable option cards, pick Omni on the model card, then resolution and ratio.
3. The live panel renders the finished clip in 20-180 seconds; failed renders auto-refund.

The connector setup is walked through in generating ads inside Claude with MCP.