At Clarity.AI, we believe content creators should have control over what’s in their video before, during, and post-production.
In a constantly changing modern world, too many brands and creators are stuck linked to brands and imagery that do not serve them anymore. The only option is for companies to remove their content all together, no matter how much effort and cost they put into it. Clarity fixes this using computer vision, offering people a way to alter visual aspects of their content at any time.
Our goal is to save brands millions. Formula One removed videos of an FTX-sponsored car with millions of views, while Bud Light was forced to scrape the internet of their latest experimental ad flop.
Clarity is made for these companies and more!
What it does
Clairy.AI employs Meta’s SAM (Segment Anything Model) to create masks over target imagery in both images and video. From here, we use E2FGVI for In-painting, both for image removal as well as image additions.
This allows companies to customize the logos and advertisements within their content based on inherently changing demographics including geo-spatial population preferences, brand value, or for frictionless special effect implementation.
Imagine adding your logo to your favorite MLB field, or making baseball fun to watch in general. Clarity can do that!
How We Built It
We used Meta AI’s segment anything model (SAM) for frame-by-frame object segmentations which we then diffuse over with E2FGVI.
Challenges We Ran Into
Initially we set a team goal of segmenting videos. After diving into the process, we realized our first approach was only suitable for imagery.
To augment our capabilities, we transitioned to a model we previously had no experience with. This felt like a relatively large hurdle to overcome and felt we would stagnate at simple image removal.
Implementation of Grounded Segment Models: We successfully utilized Grounded Segment Models to identify and isolate unwanted objects and logos in various types of images. These models demonstrated high precision and reliability across diverse contexts.