Higgsfield AI, a video AI company, has launched its first artificial intelligence (AI)-powered app for smartphones. Dubbed Diffuse, this mobile-first app is an image-to-video generator that can use a selfie to turn it into a video featuring the character. The company claims that the character it creates comes with lifelike motion. The app is being gradually rolled out to select markets and users will be able to find the app on both Android and iOS. This AI tool will likely compete with OpenAI’s Sora, which is reported to be launched later this year.

The company’s official account took to X (formerly known as Twitter) to make the announcement. Higgsfield AI introduced itself as a video AI company which is focused on creating video content for social media. This vision likely comes from the CEO of the AI firm, Alex Mashrabov, who was previously leading the generative AI division at Snap. As per the post, the company is building a foundational model from scratch for its tools which will be entirely video-focused.

In a separate post, the company also introduced Diffuse, the first mobile app built by the company, which is currently available in preview mode. “Pick out a video from the content library, select your selfie and Diffuse will generate a personalized character in that video’s style. Or use Prompt Builder to create a video from scratch using text, images or video. Diffuse offers a deep level of personalization, creative control and fine-tuning so anyone can create exactly what they want (with safeguards of course),” the post added. The app is currently available in select markets which include India, South Africa, the Philippines, Canada, and countries in Central Asia.

We were able to locate the app in the App Store, but it did not show up on the Android Play Store. This can change in the next few days as the company is gradually rolling out the app. The app currently offers video generation of 2 seconds as it is in preview mode. The AI firm says its ultimate goal is to achieve realistic, detailed and fluid video, all on a mobile device.

For this, it is building its foundational model from scratch. The video model uses transformer architectures, which are also used by OpenAI’s ChatGPT. Higgsfield AI further highlights that it was able to efficiently train its AI model on limited GPUs due to a proprietary framework developed in-house. The company has not revealed when it might release the full version of the app in public.


Shares:

Leave a Reply

Your email address will not be published. Required fields are marked *