Fal launches Agent and Agent API, model library passes 3,000
At its GenMedia conference, Fal announced its Agent and Agent API for automatic model routing, plus progress on video model post-training and a world model accelerator.
Bigger, better, and even more electric, the 2nd annual GenMedia Conference put on by @Fal last Thursday in SF was a global gathering of the growing number of really incredible builders and creative pioneers in the space. Some highlights below:👇🏽 - Fal has 3000+ models now hosted and available for inference. It was less than 50 two and a half years ago. The neolabs and established labs in generative media continue to grow and diffuse in specialty. 📈 - Fal Agent and Agent API launched and make the optimal model routing decisions for customers and devs. Building more above-the-token power and efficiency for creatives and enterprises ⏩ - @fal Research Labs moved out a new pareto frontier with a 35x performance improvement post-training an open weights video model re H3 Max. They are open for business to post-train models for any foundation model lab that seeks to commission it! 🏋️♀️ - Their new world model accelerator launched and proposes a long-horizon compute environment for world model labs and devs to explore a novel extended user experience versus the one-off generation workload. ♾️ - Iteration speed is the game changer in hollywood’s usage of gen media - not replacement of creativity and human originality. ⏩ ⏩ - best schwag bar ever. see @adamho’s timeline. ✨ - Live interactive storytelling and social experiences are now proven and possible with the latest breakthroughs in real time generation. Neo-studios are taking notice! 🕹️ - In my fireside with the head of Amazon MGM Studios, we spoke of “the Toy Story moment” that will happen soon bringing human art direction and storytelling, enhanced by frontier gen media models, to the wonderment of a mass audience. The models are now ready, and the creative artists and tools makers are making incredible progress. 🧸 - Like with LLMs, Gen Media is massively compute-bound and @fal is on the case with a multi-year commitment to build 1 GW of compute as the foundation infrastructure for the space. 🔋⚡️🏭
This post quotes another post: x.com/fal/status/2103171324831031602