stable-diffusion-webui
stable-diffusion-webui copied to clipboard
AUTOMATIC1111
[Feature Request]: Stacking or mixing variation seeds to refine a result
Is there an existing issue for this?
- [X] I have searched the existing issues and checked the recent builds/commits
What would your feature do ?
Something I like in InvokeAI is the ability to keep stacking variation seeds upon each other. I’d use this to refine a seed that was working well with ever decreasing variation strengths.
Proposed workflow
- Generate an image
- Enable a variation seed and strength
- Generate the variant
- Lock the previous variant seed and strength, then add a new variation seed on top of that, at even lower strength
- Repeat to refine the image using multiple variations at decreasing strengths.
Additional information
No response
I've been wondering about this, the resulting image would have a load of metadata, but at least these variant images would be more reproducible than inpainting or color-sketch images.
I'm also trying to figure out how to do this as this seems like extremely useful and time saving feature.
Agree, the most important feature! I have hundreds of images generated overnight. From the good ones I generate 10 variations. Then I edit endless variation seeds and hope there's finally one where the fingers are right.
Huge improvement, would Automatically generate a dozen 0.01 variations while I'm drinking a coffee.
speaking of this, i saw a post with something implemented like this:
Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: (original seed number), Size: 480x832, Model hash: ********, Variation seed: (another seed number), Variation seed strength: 0.21, Denoising strength: 0.69, Clip skip: 2, ENSD: 31337, Hires upscale: 1.15, Hires steps: 14, Hires upscaler: Latent
how can I do this in automatic1111 ? is there a plugin to install? is this what you are talking about? sorry for the newbie question but I want to learn.
The place where I copied this text from, implied the images shown were generated with A1111
