I will give you all my votes if you can help me train a model, like we had discussed earlier, that takes as input a tuple of vectors for semantic similarity between the finished image and the text prompt, the semantic similarity between the finished image and the source image (img2img), and aesthetic score of the finished image. Using the three vectors automatically generate an extended prompt, and the cfgscale, steps, seed, denoising strength, etc.
I will give you all my votes if you can help me train a model, like we had discussed earlier, that takes as input a tuple of vectors for semantic similarity between the finished image and the text prompt, the semantic similarity between the finished image and the source image (img2img), and aesthetic score of the finished image. Using the three vectors automatically generate an extended prompt, and the cfgscale, steps, seed, denoising strength, etc.