Test competing visual instructions

Why this matters

“Massive” and “tiny” describe relationships, but a generator may interpret them unpredictably. Numerical weighting syntax is provider-specific and may be ignored entirely. Start with plain language and inspect the image; do not claim a model used a weighting mechanism merely because you typed special punctuation.

Before you start

Materials are free. External tools may require an adult account, credits, regional eligibility or a paid plan. Check current terms before using them. Do not paste passwords, API keys, private messages or other people’s personal data. No model runs locally or downloads automatically on this platform.

Your illustrated field guide

Conceptual illustration, not a provider screenshot. Artwork is AI-generated; instructional labels are authored text.

Start with fictional input

Scene: a small kitten beside a large fictional dragon in a wide empty courtyard. Both subjects must be fully visible. Compare: dragon twice the kitten’s height; dragon ten times the kitten’s height; kitten larger than dragon.

Copy and adapt this prompt

Create one original illustration of the specified scene. Keep both subjects fully visible at the same distance from the camera. Change only their relative height: [chosen condition]. No text, logos, recognizable characters or extra animals.

Worked reference—not proof of correctness

A test sheet measures the approximate visible height ratio and notes perspective or cropping. If the dragon is mostly off-frame, it does not demonstrate a controlled 10:1 size relationship. The result may be a failed condition, not an artistic success.

Five practical steps

1. Fix scene and framing

   Keep viewpoint, courtyard and subject count constant.

2. Vary one relationship

   Generate the three plain-language size conditions and record model/settings.

3. Measure visibly

   Estimate each subject’s height from the image and note perspective/cropping.

4. Try one supported control

   Only if provider documentation describes weights, test that syntax separately from the language experiment.

5. Report limits

   Retain failed images and distinguish observed results from assumptions about model internals.

Test your result

• Both subjects are visible in each scored result, or cropping is explicitly recorded as a failure.
• Only the size relationship changes in the primary prompt comparison.
• Provider-specific weighting is not presented as universal or inferred from appearance alone.

Mistakes to watch for

• Treating arbitrary punctuation as a cross-provider weighting standard.
• Confusing foreground perspective with actual relative size.

Offline / no-key route

Make three simple silhouette sketches and measure the height ratios manually.

Reflect

How did perspective make the size comparison harder?

Self-reported practice is not independent assessment or a qualification. Browse all 50 guides · Choose a tool