Research protocol
Short Prompt or Detailed Prompt? A Test for Finding the Useful Middle
A transparent test of prompt length, instruction density, adherence, diversity, and the point where added detail stops helping.
The central idea
Prompt length is a poor target by itself. The useful question is how many independent decisions a prompt asks the model to honor, whether those decisions agree, and whether the result still leaves room for productive variation.
A repeatable workflow
Create length levels
Write a subject-only prompt, a compact directed prompt, and a detailed prompt from the same brief.
Hold settings stable
Use the same model, aspect ratio, output count, and review conditions for every level.
Review blind
Hide the prompt level and score adherence, visual quality, diversity, and decision usefulness.
Publish the misses
Show all scored outputs and limitations instead of selecting only attractive examples.
Worked example
A coffee-roaster brief can progress from “coffee bag on a table” to a compact version that adds audience, framing, light, and material, then to a dense version with many props and stylistic constraints. The middle version may balance direction and breathing room—but the test, not folklore, should decide.
Review checklist
- Prompt levels share one brief
- Generation settings are identical
- Reviewers do not see the condition
- All outputs and caveats are shown
Limitations
- A small test does not prove a universal rule
- Results are specific to the dated model and prompt set