Motif: Interpretation (you steer by taste, the model fills in gaps).
- Use a strict output format: single HTML only.
- Prompt once. No iterations.
- Evaluate by running the artifact and describing what the model assumed.
Motif: Collaboration (feedback loop as a creative tool).
- Generate → test → describe → request changes → regenerate.
- Change one thing per round.
- Keep prompt + parameters alongside outputs.
Motif: Rules (constraints as the artwork).
- Write constraints like a score.
- Generate multiple candidates.
- Curate the best one.
- Prompt-as-score: write prompts like performance instructions (Sol LeWitt–style) and compare outputs at different temperatures.
- Constraint tournaments: run the same rules across models (GLM 5 vs Kimi K2.5) and vote on the best artifact.
- Debug the aesthetic: treat visual glitches as signals; add one constraint to fix only one failure mode at a time.
- One randomness source: force all variation to come from a single random seed; explore how structure emerges.
- Parameter journaling: keep a small lab notebook of model, temp, max tokens, and the result.