Blog
Notes on diffusion inference, our auto-tuning inference stack and what we build with it.

Product
AI product video from one prompt: what comes out finished
What a finished product ad needs beyond a clip from a video model, and which of those parts DV4 makes from one line: shots, type, music and checks.

Engineering
Where the GPU time goes when you serve a video model
A video model in production pays for more than drawing frames: waiting, moving weights, sharing GPUs and redoing work. Where each goes, and the levers.

Product
Motion graphics from a prompt, written as code
How DV4 makes motion graphics with no template: a model writes each film as code, watches its own drafts and fixes them, and never invents your numbers.

Engineering
Why denoising steps decide what a generated video costs
A diffusion model makes a clip in many passes, not one. A plain-English look at the three things that multiply the GPU work, and the levers for each.

Product
From one line to a finished video: inside DV4
DV4 is an agentic AI platform for creators. One line in, a finished video out. Here is what happens between the prompt and the final file.

Company
Why every speed-up needs a quality check
A faster diffusion model that makes different pictures is a different model. Why we compare each speed change with the original model's output.
