search

lllyasviel / ControlNet-v1-1-nightly

Nightly release of ControlNet 1.1
4.76k stars 376 forks source link

Please help, my Controlnet Model does not work #153

Open Skyintrouble opened 2 weeks ago

Skyintrouble commented 2 weeks ago

Note: Sorry for all the mistakes I've made in this post. English is not my mothertongue Thank you in advance for all the help

I'm a newbie trying to master Stable Diffusion. I've just learned to use basic Controlnet principles and already I run into this error. I tried twitching among the Controlnet models but unfortunately had no luck. For example, when I tried to use MLSD - OpenPose - Depth Models, the results always come back like this: Screenshot 2024-11-03 224755 Screenshot 2024-11-03 224701 Screenshot 2024-11-03 223652

There're some models that actually work though, like Canny and Linear (except the Preprocessor lineart_realistic): Screenshot 2024-11-04 224048 Screenshot 2024-11-04 224025

I asked for help from some professionals and they think there's a chance that my PC system and the website-github-I-download-Controlnet-from is not suitable, like SSL certificate problem. Honestly I'm at a lost here and have no clue if it's true or how to fix it. Please help me identify and solve this error.

Some additional info to help you have a better understanding of this:

TsengSR commented 2 weeks ago

For the first two, when you already have a processed image, don't select a preprocessor. The preprocessor is there to extract this from an existing image.

Also which model are you using? The default controlnet models are for SD 1.5. For SDXL etc. you need different ones.

Skyintrouble commented 2 weeks ago

For the first two, when you already have a processed image, don't select a preprocessor. The preprocessor is there to extract this from an existing image.

Thank you for this insight. It's very refreshing, unfortunately it doesn't work with the third one. Hic.

Also which model are you using? The default controlnet models are for SD 1.5. For SDXL etc. you need different ones.

If I understand your question correctly, then I use SD AUTOMATIC1111. I actually installed Controlnet through Extensions>Available. If it's not the SD 1.5 or SDXL you mentioned, can you send me the link please?

TsengSR commented 2 weeks ago

The model means which SD Model you are using for generating images. Some are SD 1.5, others are SD 2.0, 2.1, SDXL, Flux.1 etc. Afaik each of them needs their own trained Controllnet Models. Controlnet models for SD 1.5 models do not work on SDXL and vice versa.

Skyintrouble commented 2 weeks ago

The model means which SD Model you are using for generating images. Some are SD 1.5, others are SD 2.0, 2.1, SDXL, Flux.1 etc. Afaik each of them needs their own trained Controllnet Models. Controlnet models for SD 1.5 models do not work on SDXL and vice versa.

https://github.com/AUTOMATIC1111/stable-diffusion-webui This is the SD Model I use. Seem like it's not any model you mention above?

TsengSR commented 2 weeks ago

That's the UI, not the model. The models are the *.safetensor files that range 2 to 22 GB. The controlnet models you use are for stable diffusion 1.5 (they have sd15 in their name)

Skyintrouble commented 2 weeks ago

That's the UI, not the model. The models are the *.safetensor files that range 2 to 22 GB. The controlnet models you use are for stable diffusion 1.5 (they have sd15 in their name)

Oh now I get what you're saying! We usually call the models "checkpoint" where I lived so it totally slipped my mind just now. Sorry for the confusion! I use realisticVisionV60B1_v51VAE.ckpt [139ac005d4] for all the pic in my post. I've tried other checkpoint but also have the same result.