In tests run by executives in advertising, animation, and real estate, OpenAI's Sora generated people with more realistic facial features than Runway or Pika
Financial Times :
Context & Ripple Effects
The comparison arrives shortly after OpenAI showed Sora work from a limited group of filmmakers, artists and ad agencies in its early Sora showcase. At the time, OpenAI had not set a release timetable while competing with Meta, Google and video-AI startups including Runway, as the launch race remained unsettled.
The tests matter because they focus on a commercially sensitive quality threshold: convincing human faces. Advertising, animation and real-estate users are evaluating whether video generators can produce people suitable for client-facing material.
First-order effects
- OpenAI gains a concrete quality signal for Sora among the tested users: more realistic facial features than Runway or Pika.
- Runway and Pika face a more pointed benchmark in workflows where believable human subjects are central, while the test participants have a clearer basis for tool selection.
Second-order effects
- Creative teams may concentrate trials on facial realism and other human-subject failure modes rather than judging text-to-video tools solely by novelty or prompt adherence.
- Competing video-model vendors are pressured to demonstrate comparable output quality to agencies, studios and other buyers evaluating synthetic-video workflows.
Third-order effects
- If facial realism becomes reliably differentiating, video AI competition may shift from broad demo appeal toward task-specific quality benchmarks tied to commercial production use.
- More convincing generated people also raises the stakes for provenance, consent and usage controls as these tools move closer to client-facing creative work.
The trend: Text-to-video AI is moving from showcase clips toward quality comparisons based on whether outputs meet the practical standards of professional content workflows.