EXPERIMENT 001
Can AI direct a motion identity—not just generate it?
Can AI direct a motion identity—not just generate it?
Can AI direct a motion identity—not just generate it?
A controlled test of whether a language model can build a coherent motion system across concept, art direction, and production.
A controlled test of whether a language model can build a coherent motion system across concept, art direction, and production.
WATCH THE EXPERIMENT — 08:42
DATE
July 27, 2026
DURATION
5 days
CATEGORY
Creative AI · Motion
01 / OVERVIEW
The question was direction, not generation.
The question was direction, not generation.
Generative tools are already fast at producing isolated frames. This experiment asked a harder question: can one model maintain taste, logic, and motion behavior across an entire identity system? We constrained the process to a single creative brief, a fixed visual vocabulary, and repeatable evaluation criteria.
Generative tools are already fast at producing isolated frames. This experiment asked a harder question: can one model maintain taste, logic, and motion behavior across an entire identity system? We constrained the process to a single creative brief, a fixed visual vocabulary, and repeatable evaluation criteria.
METHOD NOTE
Every output was judged on continuity, editability, and whether a human director could confidently extend the system.
02 / KEY FINDINGS
What survived contact with production.
What survived contact with production.
01
Constraints improved authorship
A fixed palette, camera grammar, and timing scale produced more original work than open-ended prompting.
02
Language became the control layer
The strongest prompts described relationships and behavior—not surface-level visual adjectives.
03
Consistency still needed a human eye
AI maintained motifs well, but rhythm and emphasis required deliberate editorial correction.
04
Reusable systems beat polished frames
The experiment became valuable when outputs could be remixed, extended, and handed to another maker.
03 / RESOURCES
Unlock the complete experiment kit.
Unlock the complete experiment kit.
Get the working files, exact prompts, source references, tool stack, and unedited field notes by email.
Project Files
AI Prompts
References
Tools Used
Field Notes
04 / CONCLUSION
AI can hold a direction—when the direction is designed first.
AI can hold a direction—when the direction is designed first.
The model was most useful as a system partner: fast enough to explore, consistent enough to extend, but dependent on clear constraints and editorial judgment. For future experiments, we recommend defining motion rules before generating frames and evaluating every output as part of a family—not as an isolated artifact.
WORKING PRINCIPLE
Design the rules. Generate the range. Edit the system.