Anyone have any tips on recreating perchance's text-to-image output on a local model? Like which exact checkpoint the generator uses, or any additional unseen parameters that go into its image generation?