Invideo, a leading platform in the video creation industry with over 25 million customers across the globe.Their customers range from individual creators to large enterprises. 

Invideo has developed intuitive, AI-driven features that simplify video creation. However, as the demand for high-quality visuals grew, the costs associated also started to increase and their existing image generation solution was no longer viable.

Deploying and managing visual content generation pipelines at scale, relying on off-the-shelf solutions can lead to performance degradation and cost increases due to the inherent inefficiencies of a one-size-fits-all approach.

As Invideo’s production load continued to increase, they faced some key challenges

• Relying on a third-party API provider for image generation meant that they were spending over $30,000 per month, which was starting to impact their investments in product development and growth
• The inference time of over 7 seconds was far too slow to appeal to their growing user base and they were facing issues with user dissatisfaction.

Simplismart conducted an in-depth analysis of Invideo's existing model pipelines and developed a plan to integrate our Simplidiffuse solution into their existing workflows. 

Simplisdiffuse uses a proprietary inference engine to deliver more performant implementations of popular Open-source models such as Stable Diffusion and CogVLM.

• The inference engine speed improvements are model-agnostic and with a suite of proprietary kernels and a universal backend optimized for ML workflows, Simplismart was able to reduce the image inference time from 7.1s to 3.5s, a 50% improvement
• The optimizations delivered across the entire pipeline resulted in much fewer compute resources required for the same workload, cutting generation costs by 97%, from $30,000 to just $903.