Thanks. So I have a bespoke python program that basically does this:
- Takes the platonic set of prompts
- Uses model specific tuning directives with LLMs to create a bunch of prompt variations so that they get a diverse set of natural language expressions to "roll" generations
But I still have to manually review each of the final image - which is pretty time-consuming. I've tried automating it using VLMs (like Qwen3-VL) but unfortunately they can miss the small details and didn't provide as much value as I was hoping.
Comments
Thanks. So I have a bespoke python program that basically does this:
- Takes the platonic set of prompts
- Uses model specific tuning directives with LLMs to create a bunch of prompt variations so that they get a diverse set of natural language expressions to "roll" generations
But I still have to manually review each of the final image - which is pretty time-consuming. I've tried automating it using VLMs (like Qwen3-VL) but unfortunately they can miss the small details and didn't provide as much value as I was hoping.