[Feature]: add extra noise

[RGX650]

Feature description

Add noisy latent to ExtraNoiseParams for callback
AUTOMATIC1111/stable-diffusion-webui#12856

Add extra noise callback
AUTOMATIC1111/stable-diffusion-webui#12616

Add extra noise param for img2img operations
AUTOMATIC1111/stable-diffusion-webui#12564

[catboxanon]
https://gist.github.com/catboxanon)/[zz_extra_noise_mask.py]
https://gist.github.com/catboxanon/69ce64e0389fa803d26dc59bb444af53

Version Platform Description

No response

[vladmandic]

i'm ok with this, but can you submit actual PR?

[RGX650]

thanks for your interest and reply!

I am not a proper dev actually, git related methodologies an subtilities are still a bit obscure to me...
so please don't be mad at me.

i kind of manage to start implementing it through my fork
https://github.com/RGX650/sd-webui-bmab-SDNEXT/blob/main/README.md
https://github.com/RGX650/sd-webui-bmab-SDNEXT

It is a fork of @portu-sim BMAB extention working with A1111
https://github.com/portu-sim/sd-webui-bmab

Adding A1111's RNG
https://github.com/RGX650/sd-webui-bmab-SDNEXT/tree/main/sd_bmab/external/rng

Adding extra-noise callbacks
RGX650/sd-webui-bmab-SDNEXT@09d9aec
https://github.com/RGX650/sd-webui-bmab-SDNEXT/blob/main/sd_bmab/sd_override/__init__.py

shared.opts.add_option('img2img_extra_noise'
shared.opts.add_option('randn_source'
https://github.com/RGX650/sd-webui-bmab-SDNEXT/blob/main/sd_bmab/ui.py
synch extra-noise value bmab/webui? #1172

class StableDiffusionProcessingTxt2ImgOv(StableDiffusionProcessingTxt2Img)
https://github.com/RGX650/sd-webui-bmab-SDNEXT/blob/main/sd_bmab/sd_override/txt2img.py

class KDiffusionSamplerBMAB(KDiffusionSampler)
https://github.com/RGX650/sd-webui-bmab-SDNEXT/blob/main/sd_bmab/sd_override/samper.py

[vladmandic]

I am not a proper dev actually, git related methodologies an subtilities are still a bit obscure to me...
so please don't be mad at me.

mad? far from it - i appreciate the work!

i'll check out your repo soon, really want to wrap-up work on new control module.
also, i will need to implement callback for diffusers backend (where callbacks work quite differently) - but it should be doable.
with more and more features in diffusers backend, i think there will be even more value there.

[vladmandic]

sorry for the delay, last month has been pretty crazy.
anyhow, img2img_extra_noise is now added for backend:original k-diffusers. if there is sufficient demand, it can be added to remaining compvis samplers (ddim,unipc). backend:diffusers will need separate implementation.