Use the named model
Start where it was tested.
Each prompt names the image model it was built around, so you are not guessing how it behaved in the first place.
Image Models
Prompts by image model
Start with the model you are using, then pick a prompt that already has the camera flaws AI tends to smooth away: harsh flash, soft room clutter, skin texture, bad crops, and enough real-life mess to keep the result from looking staged.
New model pages have to earn their space with tested prompts, not name-dropping. Wan, GPT Image, and Pro matter here only when they solve the same old problems: clean skin, perfect rooms, fake flash, stiff hands, and prompts that fall apart the second you paste them.
48
live prompts
1
live image model
3
models queued
Live Model
Open the model with tested prompts.
Nano Banana 2 is live now. Wan 2.7, GPT Image 1.5, and Nano Banana Pro stay queued until they have enough tested prompt pages to be useful.
Keep one prompt page
No cloned prompt sludge.
A prompt keeps one permanent page. The model label tells you where it was tested without turning the site into duplicate filler.
Switch carefully
Carry the ugly details over.
If you move a prompt to another model, keep the flash, clutter, pores, framing, and room evidence. That is usually what keeps it real.
Compare honestly
The model changes the cleanup.
Wan may hold reference detail better. GPT Image may follow a stricter setup. Nano Banana 2 may get you there faster. The point is knowing the tradeoff before you paste.
Queued Models
Next up for prompt tests.
These models are on the map, but they do not get a live model page until the prompts are tested enough to help.
Nano Banana Pro
complex prompt adherence · higher-fidelity production assets · structured prompt tests
Wan 2.7
reference-heavy image work · collection batch testing · garment and room texture
GPT Image 1.5
instruction adherence · complex scene setup · cross-model comparison