posted on X
7. Make it more varied After Claude gives you the first concepts, paste this: These concepts are too similar. Make them more visually different. Change: 1. The hero object 2. The camera angle 3. The emotion 4. The lighting 5. The background 6. The composition 7. The visual metaphor 8. The level of realism Give me 10 new versions that feel like completely different creative directions.
View original tweet on X →
After an AI returns too-similar visual concepts, this follow-up prompt forces it to diverge across eight distinct creative dimensions.
Prompt
These concepts are too similar.
Make them more visually different.
Change:
1. The hero object
2. The camera angle
3. The emotion
4. The lighting
5. The background
6. The composition
7. The visual metaphor
8. The level of realism
Give me 10 new versions that feel like completely different creative directions.Why it works
LLMs tend to cluster around a central interpretation of a brief, producing variations that are superficially different but structurally the same. By explicitly naming eight independent visual axes, this prompt forces the model to treat each dimension as a separate degree of freedom — making it much harder to recycle the same underlying concept.
The axes are well-chosen because they span orthogonal aspects of visual design: subject matter (hero object), spatial perspective (camera angle, composition), narrative tone (emotion, visual metaphor), technical render style (lighting, realism), and context (background). Changing all eight simultaneously virtually guarantees distinct creative directions rather than palette swaps.
Asking for 10 versions compounds the effect — with eight axes and 10 outputs, the model is statistically pressured to distribute across the possibility space rather than converge on a local optimum.
When to use
- •When an AI's first batch of visual concepts or ad ideas feel repetitive or too on-brand/safe
- •During early creative concepting for campaigns, thumbnails, or product visuals where range and diversity matter
- •When briefing an image-generation model (Midjourney, DALL-E) and you need varied prompt seeds before committing to a direction
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.