How HeyGen cut image generation costs by 50% without the infrastructure overhead

Images for HeyGen play a critical role in their workflow. High-quality images help users visualize their final output far earlier in the process. A user might start with an initial frame or build out a storyboard before committing to a full render, iterating quickly and confidently along the way.

To make that workflow viable at scale, HeyGen needed image generation that was both low-latency and affordable. The two options on the table were managed providers and self-hosting open-source models like FLUX. Managed providers offered convenience but came with costs that didn't hold up at production scale. Self-hosting offered theoretical savings, but would have required a significant engineering investment to build and maintain the supporting infrastructure. HeyGen didn't want to pull engineering resources away from its core product.

The team needed a third option: managed simplicity at self-hosted economics.

Runware's API was highly compatible with their existing workflows, making migration straightforward and requiring minimal engineering effort to complete.

Runware enabled HeyGen to:

The switch struck the balance HeyGen was looking for: the reliability and ease of a managed service, without the cost structure that had made other providers unworkable at scale.

The impact was immediate and measurable. Overall cost per image dropped by over 50%, which changed how the team approached product development. Features that had previously felt too expensive to build and release at scale became viable. HeyGen could now ship new image-based capabilities without running the numbers every time.

Key outcomes:

Beyond the cost savings, Runware has proven consistently reliable under real production conditions. HeyGen has experienced no major issues since switching, and the platform has become a core part of their infrastructure going forward, serving as their primary solution for open-source image models.