kyndcreate

KyndCreate/Blog

The KyndCreate blog

Notes from the director’s chair.

What we learn building an AI film studio: which models direct well, what they cost, where they break. Real frames, measured numbers, and a plain note whenever results are still coming in.

Four versions of the same close-up of a woman speaking on a clifftop with a red lighthouse behind her, labelled Draft, Fast, Omni and Max with their credits

Kynd’s four quality levels: Draft, Fast, Omni and Max, side by side

One close-up and one wide shot, at every Kynd quality level. Watch them side by side, see the faces up close, and find out what each level costs and which one suits your shot.

See the comparison →
A lightning bolt strikes a prairie between a farmhouse and a TV reporter, one of GLM-5.3 Flash's opening frames

Which AI model is best for video? Kimi K3, GLM-5.3 and DeepSeek V4 direct the same three films

We had three language models direct the same three short films, drew every picture with the same image models, and let the models judge each other blind, with Claude as an outside judge. In the first run each model won one film and DeepSeek V4.1 Flash edged it overall; a re-run with a KyndCreate fix is under way. Here are the frames, the scores so far, the speed and the cost.

Read the comparison →

Notes on running AI on your own Mac (local models, memory, speed) are on the Kynd Agent blog.