Archive

The first courtroom answered.

Shipped, or merely announced.

Four releases in seven days. The question worth asking is not which is most impressive. It is which one you can put to work this afternoon, and the answers are further apart than the headlines suggest.

Two clips, one room.

Generated shots arrive as strangers. Each one is a separate roll, and nothing in the model obliges the second clip to believe it is standing in the same room as the first. Continuity has to be written before it can be watched.

Write the whole world.

FLUX 3 is in early access, not general release. Its larger idea is a useful creator habit: write image, motion, sound, and action as one coherent world.

When your taste should speak.

Midjourney V8.2 landed on July 24 with a personalization system that reads your rating history more closely. That makes your own taste a louder ingredient in every render, and it makes knowing when to quiet it part of the craft.

Five models on the bench.

Leaderboards rank models in general, which is a question nobody's workflow actually asks. A bench ranks them by the job. The five seats in our current rotation, and what each model must keep doing to hold its place.

Keep the judgment gate.

Google's first ATLAS report studied 15 million interactions and found that fewer than one in ten workplace uses fully automate a task. The useful pattern is collaboration with a named handoff.

A two-speed route for repeatable AI work

Google's new Flash pair turns model choice into a routing decision: one lane for difficult planning and multimodal judgment, another for repeated extraction, classification, and subagent work.

The ten-second review: judging AI images at named points