Practical guide
The Change-One-Thing Method for Better AI Image Variations
Learn more from every batch by changing one major creative variable while keeping the rest of the prompt stable and reviewable.
The central idea
When composition, palette, setting, medium, and subject all change together, the batch cannot explain why one result works. A single-variable test creates comparable options and a clear next move.
A repeatable workflow
Save a baseline
Keep one prompt and output as the control so every later choice has a visible reference point.
Choose the decision
Select the highest-uncertainty variable: composition, palette, setting, light, material, or medium.
Write bounded variants
Name four distinct values for that variable and keep the rest of the prompt identical.
Promote the winner
Lock the chosen value into the next baseline before testing a new variable.
Worked example
For a picnic campaign, keep the people, products, time of day, and relaxed tone stable. Test only centered group, wide negative space, overhead arrangement, and foreground-framed compositions. The review can now isolate which composition supports copy and product recognition.
Review checklist
- A saved baseline exists
- One major variable is under test
- Variants are meaningfully different
- The winner becomes the next control
Limitations
- Generative randomness means even controlled tests are not laboratory-perfect
- Some variables interact and require a later combined test