ai& launched ai& video, an AI video generation service that turns text, photos, and existing video clips into new footage. Creators and companies can describe a scene in Japanese, bring a single photo to life, or continue an existing clip, generating video in as little as 12 seconds* on ai&’s own infrastructure in Japan.

Through ai& inference, its AI inference platform, ai& has made AI models available for applications ranging from language processing to scientific research. ai& video extends this offering to video generation.

Users can describe who, where, or what appears in a scene, where it takes place, and what happens, all in plain Japanese, and have the entire scene generated. They can also animate a photograph, using its subject and lighting as a starting point, or continue an existing video clip. Fast generation allows users to experiment with camera movements, compositions, and alternative creative directions. Businesses can also use the service to visualize concepts and evaluate options early in a project. On supported models, ai& video also generates audio alongside the video.

Marketing Technology News: MarTech Interview with Mark Listes, CEO @ Pendulum Intelligence

ai& video is available through two interfaces:
Web browser: Enter text or upload materials to generate videos directly in your browser.
API: Integrate video generation into existing production tools or your own applications.
Pricing is pay-as-you-go and varies by model, with no monthly subscription required.

For enterprise customers, ai& video supports additional training using characters, footage, and other assets they have rights to use. This produces a dedicated model that reflects the company’s visual identity and style and is available exclusively to that company. Additional training takes place in a dedicated environment that other users cannot access. Materials uploaded for standard video generation are never used to train models.

All video generation on ai& video runs on ai&’s own infrastructure in Japan. An in-house moderation system developed in-house helps prevent inappropriate generation and uses that could infringe intellectual property rights.

Video generation is computationally intensive. Beyond producing high-quality individual frames, the model must keep people and objects moving coherently over time. Even a few seconds of footage can be costly to generate. Making generation faster requires maximizing the performance of the compute platform for each model. ai& optimizes inference for specific combinations of models and chips, developing runtime software that takes full advantage of NVIDIA GPUs and processors from other vendors. This combination of hardware selection and in-house optimization delivers high throughput at low cost.

Marketing Technology News: How MarTech Is Enabling Autonomous Brand Engagement Across Channels?

Built on this inference technology, ai& video generates video in as little as 12 seconds* by processing computations efficiently. Reducing the time and cost of each generation allows creators to experiment with more ideas.

“Language, drug discovery research, video: the form inference takes changes with the model and the use. The computation each one needs, and the way to bring out its performance, are different too.

In video work, the next idea comes after you see the first clip. You want to try a different movement or change the composition. We want to build an environment where waiting time and cost never make you stop.

We build everything from the infrastructure to how models run because we want to connect the progress of AI to what the people using it can do. We want to grow ai& video into a place where creators and companies can try new forms of expression together,” said Shimpei Hara, co-founder and president, ai&.

ai& will continue to add support for new video generation models while improving generation speed and processing efficiency. As AI expands into new forms of information and expression, ai& will also broaden the range of models and domains available through ai& inference.