Is Kandinsky-2 is better than Stable Diffusion?

Even more free image models? Yes, please! Welcome to Kandinsky-2 a sort of mix between Dalle-2 and Latent Diffusion. Make images from text, do image inpainting, mix images and more - all for free on your own PC :)

== Install ==
conda create --name kandinsky-2 python=3.10
conda activate kandinsky-2
conda install -c conda-forge notebook ipywidgets opencv
pip install protobuf==3.20.0

== Links! ==

== Stable Diffusion ==
* Interested in adding things to your AI Art? Try these!

[thomassynths]

It's important to distinguish the people living in bad countries from their rulers. These are just everyday people making fantastic software.

[sharifhamza]

Great model. Just needs a webui like automatic 1111. Thanks for the video I love it.

[kariannecrysler]

Shhh, I’m sneaking away from my yard work and planting prep. I I had to say hey to my favorite whiskery nerd lol. Missing running into you ✌️💗🤗

[mich_elle_x]

Kandinsky model image generation has been available in Russia for several mouths through the smartphone app by Sber (a Russian state bank that since some time ago is also a tech company).

[Vigilence]

It’s much better then the base version, but I would like to connect Lora’s, textual inversions, mix models and use negative prompts.

[banzai]

I like that. Keeping them honest. Looks good, downloading

[JiraMelaj]

I'm interested in how Kadinsky-2 stacks up against IF by StabilitAI's DeepFloyd AI Lab! :P

[synthoelectro]

I think Kandinsky has more flavour for us. I've noticed a glaring problem with Stable Diffusion, stuff starts to look the same, way too much. Faces appear the same, just stuff in general is bland, but Kandinsky always had something more. I've messed with it in the past and in the early days it was a real mess, but look at it now, very impressive.

[wakariah]

wonder if some kind of cross attention feature could get this working on a 3070 8gb

[phizc]

Kadinsky 2.2 is out now. Ai Flux made a video about it today, but he's too afraid to run russian code on his computer. Out of curiosity, did you get a nasty infection out of this? 😂

[rendezk]

Been using this through Stablecog and Dreamlike - both solid options for ease I would say!

[morizanova]

Kadinsky seem offer rich color and nice contrast so meanwhile SD more mature in prompt follow part. I wonder if we can use both world . Bring SD text2img result and doing img2img in Kardinsky them bring back again ij SD for inpainting and upscale

[mekurri]

Kandinsky 2.1 is mad fun, god i hope deforum colab adds it asap so we can do animations

[nicolastines]

Thanks for your always clear explanations but why do you use SD1.5 to compare ? 2.1 would be better no ?

[thailad_]

Someone wanna tell him that Voldy is Russian too?

[sambas]

hi Nerdy can you do a video tutorial on how to install Kandinsky 3.1 on Comfyui with a diffuser support node?Comunity totally boycotted Kaninsky only because it is russian but it is very good model on par with a refined sdxl one. It is very sad this kind of things happen even in art

[synthoelectro]

hasn't google taken away the ability to use a local UI, they mentioned something about this in April.

[Graydient_AI]

To be fair, the Stable Diffusion user isn't sticking to the base model. And really that's what the Kadinsky advocates should focus on -- how do you beat the largest fine-tune dreambooth repository of every popular subject, and why would those model makers come over?

[fluffsquirrel]

Vlad just came out. Now there's a new one?! Geez!

[vheypreexa]

Testing it against an almost 1 year old Stable Diffusion model. It better be better. 😛