Open DeXtmL opened 1 year ago
Just for image generation, you can refer to diffusers. They provide friendly interface to use diffusion models in few lines. But the 3D model conditioned on the text can be very different from the image from the same text. So I guess it is hard to set the expectation based on generated image.
Need: giving a prompt, a seed number, select IF or SD, and generate several images as previews before the training. This is important to confirm the prompt is ideal (even if there are RNGs in play).