GPT Image 1.5
Balanced illustration and reliable prompt fidelity.
Five image minds. One brief. Ten rounds. No mercy.
This is not a leaderboard. It is a craft table.
Give the same strange brief to five image models. Look for taste, coherence, material intelligence, and the moments where a model surprises us.
The format
10 rounds
× 5 models
× 1 shared prompt
The first seven rounds are loaded here. The deck stays useful while the lab keeps cooking.
Balanced illustration and reliable prompt fidelity.
Newest OpenAI contender. Clean composition with richer detail.
Composition-first. Strong at layered concepts and legible worlds.
Fast, fluent, and unusually willing to make a complete image.
High-energy visual improvisation. Loves the bold gesture.
rounds loaded
active deck images
models in the mix
archive images retained
| Provider | Round 01 | Round 02 | Round 03 |
|---|---|---|---|
| OpenAI 1.5 | |||
| OpenAI 2 | |||
| Gemini 3 Pro | |||
| Gemini 3.1 Flash | |||
| Grok Imagine |
Bars are visual pacing, not a scientific score. Rounds 01 and 04–07 have five contenders; Rounds 02–03 retain the three loaded provider sets.
A vivid sunrise over layered mountains. One prompt, five model signatures, and no scientific score pretending to be taste.





The best image makes the brief feel inevitable—not merely obeyed.
One humble ingredient becomes fine dining, a surreal landscape, and a tattoo emblem in one frame.



Gemini Pro understood the assignment as a single visual sentence.
Glass, velvet, smoke, polished metal, and melting ice. No collage. No excuses.



The boundary between food and object gets deliciously unstable.
A dragon-fruit taco, a rainy neon market, and one shared brief. Can five models invent a world without losing the food?





The brief works when the food stays food inside the fantasy.
Six iconic designs on one sheet: panther, eagle, dagger, rose, swallow, and heart. Bold linework is the whole point.





A flash sheet is a test of subtraction.
Glass, chrome, liquid, and hard specular light. No cinematic fog to hide the physics.





The camera is honest when the materials disagree correctly.
Sugar-glass vaults, pastry buttresses, ramen bridges, chocolate terraces, and a waterfall of honey. One building. One world.





We asked for surreal. The winner still needs load-bearing logic.
| Axis | Question | What wins |
|---|---|---|
| Brief fidelity | Did it actually take the challenge? | Specificity without clutter |
| Visual intelligence | Does the composition make a decision? | A clear point of view |
| Material truth | Do surfaces behave convincingly? | Texture you can almost feel |
| Surprise | Did it find a door we missed? | Strangeness with control |
No single model wins every kitchen.
OpenAI 2 brings detail, Gemini Pro builds the most composed worlds, Flash reaches a complete answer quickly, and Grok keeps finding the sideways idea. The interesting part is the disagreement.
The real winner is the brief that exposes a difference.
Ten rounds. Seven are now loaded. One public lab notebook for model taste.
MasterAI.jimmys.tools