itch.io

I figured it out. The improvement you are able to see is because nvidia fixed the driver compatibility issue with CUDA 11.8 on the 4k series of cards. Everyone using a previous generation is already running at maximum performance.

The improvement i'm seeing is by 1.5 seconds. Sometimes slower depending on the CUDA dlls version. 

I'm testing this for sure. Where did you find version 2.3.0 

I'm not sure I understand. What's not working ? The software or the update ? Join me on discord, I'd be glad to help.

The release has been updated. Make sure you run "update_experimental.bat" from your invokeai folder.

Chose option 1 and press enter. The update process will begin.After not too long your will receive a message informing you the update was successful.

You may now launch InvokeAi using the launcher (do not pay attention if the launcher says version 2.3.2, it's just a bug). What matters is the version displayed in the Web UI, in the upper left corner.Do not forget to post your reviews, questions and your own creations here.

For models please visit Civitai. Feel free to start using InvokeAI with a powerful model:

  Crystal-Clear_v1 | Stable Diffusion Checkpoint | Civitai

Sometimes it depends on the checkpoints you have installed. I recommend updating to version 2.3.4 since it is also faster. You should convert checkpoint to diffusers for maximum speed, it literally halves the rendering time.

That's because this is a standalone version "by Sunija".  So I guess he will update it when he can. You can download the other one as well, I ran both for a while.

Update 2.3.4 is out !! Now with LORA and LYCORIS support.

Get it here: Release InvokeAI Version 2.3.4 - A Stable Diffusion Toolkit · invoke-ai/InvokeAI (github.com)

it's called k_dpmpp_2   

"k" means Karras and "pp" means plus plus

Probably Sunija is working on his launcher.

Version 2.3.3 is out, change notes: Release InvokeAI Version 2.3.3 - A Stable Diffusion Toolkit · invoke-ai/InvokeAI (github.com)

Worst case scenario, use your cpu, you'll be able to render mostly anything, it will just take a long time...

Do you happen to know if i add an extra card, will it help with the  rendering time ? or it will just render a 2nd image on the side ?

So, today marks it one week since i downloaded this and I got to say, i only slept 12 hours in one week. I've been stuck to my computer ever since. After 3 days of using the cpu to render a picture in 55minutes, I went and I bought a 3060 without even looking at the money(lucky the card was 150$ off on sale). My only regret is that they didn't have a 3070 in stock. I just want things to load faster!! I'm considering of adding another gpu, maybe even an older one as long as it runs the latest cuda drivers.

One question: i would like to compile my own sampler, where could i find the source code for the existing samplers ?

Not to forget, a huge THANK YOU bro, for making this software real.