Hi! In this article, I'm suggesting things for the next Imagen models. I'm writing it here because I don't know where to suggest changes for Imagen models, but if you know of where to do that, please tell me. Also, if you agree with my wishes, you can say so in your comments!

Wishes for the next Imagen models

• Additional Resource Compatibility: I would like the Imagen model to be compatible with LoRAs, LyCORISes, embeddings, DoRAs, etc. It would allow for enhancements in the generation results using it.

• Ability to Create Perfectly Centered Images: I would like it to be able to generate perfectly centered/symmetrical images. This would be great for when you're making a front view character template, so you don't have to move the symmetry ruler.

• No Artifacts on Complicated Compositions: I wish it won't frequently make artifacts on complicated compositions (unless stated in the positive prompt). This would be great especially for images with small faces, text rendering, and thin structures.

• Trained on Images from Many Diverse Sources: I wish it'll be trained on images from more diverse sources, like Rule 34, Danbooru, E621, Sketchers United, DeviantArt, etc. I'm not a fan of XXX-rated content, but I would especially like it to be trained on images from the Rule 34 booru, since it's a booru where I see the most diverse types of content, including lesser-known content, images/concepts that may not be seen on most other sites, and creative, never-before-seen tags.

• Image-to-Image: I'd like it to be able to used for img2img, not only txt2img. It allows for enhancing an image's quality, changing the image's subjects, etc.

• More Settings: I'd like it to have more settings, like guidance scale, number of steps, etc.

Problems

• Artifacts on Complicated Compositions: It may generate artifacts on complicated compositions, especially on small faces, text rendering, and thin structures.

• Imperfectly Centered Images: It may struggle with generating a perfectly centered image, which is especially unfortunate when you want to edit a front-view image you'd like to use for a reference sheet or when you wanna use a front-view character template made using the model (as you'd have to move your symmetrical ruler to make it symmetrical).