The Pulse
ai& Launches Video Generation on Its Own Japan-Based Stack
ai& launched ai& video on October 22, offering video generation from Japanese text, photos and existing clips through a browser and API. The service runs on ai&'s infrastructure in Japan and lets enterprise customers train private models on
AI.info Team · Written with AI · fact-checked ·
ai& launched its video-generation service on October 22, bringing text-to-video, photo animation and clip continuation onto the company’s own infrastructure in Japan. The service, ai& video, accepts scene descriptions in Japanese and is available both through a web browser and an API for integration into production tools and applications.
ai& says generation can take as little as 12 seconds under specified conditions. The company’s announcement cautions that actual times vary with the model, settings and video length; it does not give a single expected runtime for all jobs.
ai& video accepts text, photos and existing clips
Users can describe a scene—including who or what appears, where it takes place and what happens—or provide a photo as the starting point for an animated clip. They can also generate a continuation of existing video. Supported models can produce audio alongside video, while rapid iterations are intended to let users compare camera movement, composition and other creative choices.
The company also pitches the service for early-stage business work, such as visualizing a concept or comparing creative directions. Browser access provides a direct way to create clips; the API lets developers call video generation from their own applications or production systems. Pricing is pay-as-you-go and varies by model, with no monthly subscription required, according to ai&.
Enterprise customers can build private visual models
For business customers, ai& video supports additional training on characters, footage and other assets the customer has rights to use. The result is a dedicated model intended to reflect that company’s visual style and available only to it. The company says this additional training takes place in an environment inaccessible to other users.
Materials uploaded for ordinary video generation are not used to train models, ai& says. That distinction matters for organizations that want to generate content from existing brand materials but do not want routine uploads to become training data. The announcement does not specify which rights checks customers must complete or describe the service’s process for verifying asset ownership.
ai& ties speed to model-and-chip tuning
ai& says it runs all video generation on its own infrastructure in Japan and tunes inference for particular combinations of models and chips. Its announcement describes runtime software designed to use NVIDIA GPUs as well as processors from other vendors. Video models must keep people and objects moving coherently across frames, a computational burden the company says makes even short clips costly to generate.
Shimpei Hara, ai& co-founder and president, described the service as a tool for repeated creative trials: “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.”
The company also says ai& video includes an in-house moderation system intended to help prevent inappropriate generation and uses that could infringe intellectual-property rights. Its announcement offers no performance figures for moderation or details on how those safeguards handle disputed rights claims.
The 12-second figure comes with conditions
The fastest generation time is a measurement under specified conditions, not a timing guarantee for every model or request. ai& says duration varies with the model, settings and clip length, so the headline speed should be read as a best-case claim rather than a universal turnaround time.
ai& plans to add support for more video-generation models and improve generation speed and processing efficiency. The service’s central pitch is a combination of Japanese-language creation, enterprise-specific models and generation on ai&’s own infrastructure; the company has not named a launch price per model in its announcement.