Features: the models, voices, and assistant inside Overlume
Every frame starts with a frontier model.
The image and video models behind your ads, kept current. Swipe through. Each sample is that model's own render.
Seedance 2.5
Video model




Then a voice sells it.
Five named reads with real character, ready to sit over any cut. Press one to hear it.
One studio.
Full control.
Type a prompt, pick a model, and the render lands in the same bar. Three prompts, three models, three finished frames.

A red fox on a mossy rock in a misty Nordic forest at dawn
Nano Banana 2

A hot air balloon over pink salt lakes at golden hour
Soul 2

A snow leopard on a Himalayan cliff at blue hour, falling snow
Z Image
Two live ad formats
Skeleton Hook and Slideshow are live today. The rest are previews, not selectable yet.
Next format
The date we are working towards for the next format.

Skeleton Hook
Opens on the problem, then proves the product fixes it.
- ?No preview yet
Slideshow
A still-image carousel of your product, built one slide at a time.
- ?No preview yet
UGC ContentSoon
Everyday creator-style clips built around what you sell.
- ?No preview yet
More HooksSoon
Opening beats built to hold the first three seconds.
Live, while you film
Overlume runs Lucy, Decart's realtime engine, inside the studio. Film in a set that isn't there: swap the scene, the outfit, even your face, while the camera rolls.
Rolling out now in the dashboard. Sessions run as short prepaid blocks, the seconds you don't use come back, and the pricing page has the numbers. Live output may include Decart's watermark, and Lucy has a public demo at lucy.decart.ai.
Ask the studio.
Pulled the hook onto Signal Coral and set it in Clash Display at 2x weight, so it reads clean over the video. Applied it from your kit:
Lumo Coffee
lumo.coffee
And if you ever get stuck, I'm right here.