I’m trying to create a pickaxe that will use Flux to generate an image with a niche-based affirmation. Every image I try to have it create results in at least one typo on the final result.
Is there a combination or prompt tweak you recommend for getting this at least somewhat better? I don’t expect 100% perfection every time, but I would love at least one done right.
I’ve tried GPT 3.5 Turbo, 4o, and Claude 3.5 sonnet
I would use Recraft V3 if you are going to be using Text in the image
Even the best AI models are not flawless when it comes to text, with an error rate of around 1 in 4, resulting in occasional typos in generated images.
I recently began diving into Flux over on Replicate. I haven’t tried with text embedded into the image. For now, I can confidently say that if you want consistent and stable results, don’t use Flux Schnell or OpenAI’s DALL.E. You might get good results with Flux Dev LoRa, however. Worth a try!
I’ll try to update later