>>108655286
How does Nano banana pro handle this?

>>108655301
They all failed.

>>108655318
what about grok

>>108655286
i don't think any of the image generators are really that good at extremely detailed refinements yet. we're probably still a ways off from there.

>>108655443
It could be done if there was a model similar to claude that used multiple agents and perhaps even multiple models, as well as a central "commander" model.
You would have one drawing expert that can generate realistic images but cannot understand things, then you would have one model that can understand images.
The second model would convert both chess images into positions and mark the regions in the generated image that are wrong, then send feedback to the central model which would again prompt the drawing model to fix these specific regions ...

worked for me

Worked on my machine.

>>108655286
>It's censored as fuck and only gives you like 3-4 free gens
Nah, fuck this. Why would I use it over Nano Banana or Grok?

>>108655343
They ALL failed.

>>108656495
They all failed today.

>>108655286
What you're asking for is technical drawing, which demands a level of precision that visual models simply lack right now. They currently cannot produce valid blueprints, schematics, many kinds of UIs, etc.

File: my-favorite-scene-from-i-robot-v.jpg (80.1 KB)

80.1 KB JPG

>>108655286