search

Acly / krita-ai-diffusion

Streamlined interface for generating images with AI in Krita. Inpaint and outpaint with optional text prompt, no tweaking required.
https://www.interstice.cloud
GNU General Public License v3.0
6.47k stars 312 forks source link

Live Drawing Issues Help #833

Closed DDZ0920 closed 2 months ago

DDZ0920 commented 3 months ago

Hey everyone, This plugin is absolutely amazing and super useful! Huge thanks to the developer! However, I've run into a few issues:

1️⃣Using the 1.5 model + LoRA on a 512x512 canvas works smoothly. However, when I switch to 1024, it needs to "load 1 new model" with every stroke. This significantly slows down the sync rate, making the generation time longer and preventing real-time display.

2️⃣When using the XL model, whether on 512 or 1024, it always shows "loading 1 new model." This really ruins the real-time experience.

Does anyone know why this is happening? Any solutions? Thanks a lot!

企业微信截图_457b4491-ffc2-48f7-974c-e60d8d0646b4 企业微信截图_f5525a11-1c25-489e-8ec6-eb281ca3e912
DDZ0920 commented 3 months ago

What I mean is, with every stroke, it needs to load this "model" again, which is a huge waste of time. What exactly is it, and is there any solution?

企业微信截图_a700a95e-b7a4-48d8-a91d-46b1180c42bb
Acly commented 3 months ago

This usually means you don't have enough VRAM to keep everything on the GPU, so it has to load/unload for different parts of the process. And yes that ruins performance.

You can try to override ComfyUI's detected settings by passing --highvram for example. But if you don't actually have enough VRAM, it will just crash.