Research at Creatify
Creatify Labs is a research group building the models behind lifelike AI video. We publish what we learn, document what our systems can and cannot do, and hold the work to one bar: realism you cannot tell from real.
Founded by researchers and engineers from Meta, Snap, FAIR, ICML, Test-of-Time Award
"Our bar is simple: if a viewer can tell it is AI, the model is not finished."
Realism is a research problem.
Most systems animate a mouth. The harder problem is everything else: gaze, micro-expression, head motion, and holding identity and fine detail stable over time. That is where generated video still breaks, and where perceived realism is won or lost.
We treat realism as a measurable target. We evaluate models on identical inputs, track the specific failure modes that give synthesis away, and report the results, including the cases where our own models fall short.
We build models, not features. We document training data and limitations, and we would rather publish a weakness than hide it. Trust in synthetic media has to be earned in the open.
Reports and model notes
Video you can't tell from real.
We build the models, publish what we learn, and hold the work to one bar.
Explore the models, or contact us.
