[Feature request] Ability to Select Correct / Good Result from generated images to generate more similar

[sachingulwane1]

it will of great help if it is possible to add Ability to Select Correct / Good Result from generated images to generate more similar images, similar to that image, similar to what midjourney have.

it can be done using img2img but it will be great to have it in text2image, coz img2img is kind of more bound to that particular image, but text2image kind of creates more similar concepts.

Ref. something similar to the V1 V2 V3 V4

https://i0.wp.com/img.olhardigital.com.br/wp-content/uploads/2022/08/midjourney_5.jpg?resize=708%2C420&ssl=1

Also, Really thanks and appreciation for the efforts of building such a wonder full tool.

[pomee4]

What is the difference from sending it to img2img?

[sachingulwane1]

Image to image is more bound to the image you provide, this is for text 2 image, where its not bound to particular image but give more variations, as per my understanding it is kind of way of giving feedback to AI what exactly you are looking for. so it can stick to that concept for next generations.

[1Garrett2010]

What is the difference from sending it to img2img?

One possible benefit is to have consistent text effects between different icons, as I have tried to explain in this thread:

#2074

[djkacevedo]

The extra button next to seed, kinda does this, it lets you vary the generation with a second seed (can be random), which basically does exactly what I think you want, as long as you leave the original seed the same.

[ClashSAN]

yes, set it to generate random variations. Use variation seed -1, and a very low strength. You may even want to add an extra zero in the ui-config settings for even lower strength. This helps alot for dreambooth models when the variations are too drastic.

[ClashSAN]

@sachingulwane1 you wanted this https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#variations

[sachingulwane1]

Yes, kind of, thanks for the info.