unknown model architecture: '' and SEGV while loading the model sdxl_lightning_4step.q8_0.gguf

[yurivict]

koboldcpp fails to load this model:

llama_model_loader: loaded meta data with 0 key-value pairs and 2641 tensors from /disk-samsung/llama-cpp-experiments/models/sdxl_lightning_4step.q8_0.gguf (version GGUF V3 (latest))
llama_model_load: error loading model: error loading model architecture: unknown model architecture: ''
llama_load_model_from_file: failed to load model
Segmentation fault

Version: 1.72 clang-18 FreeBSD 14.1

[LostRuins]

You need to load stable diffusion models with --sdmodel instead of --model (or using the GUI, select it as "Image Model", not "Model")

[yurivict]

There's no such field "Image Model". There is the field "Stable Diffusion Model" on the "Image Gen" tab. But this model fails later:

Warning: KCPP text generation not initialized!

Output: 

This model is supposed to generate images.

[LostRuins]

There is the field "Stable Diffusion Model" on the "Image Gen" tab.

Yes that is correct.

I am confused. Do you not want to generate images? You don't prompt for text when you are generating images.

Instead, click and "Add Img" button and then select "Custom Prompt" and type your image prompt.

Alternatively, KoboldCpp also comes with an image gen SD UI http://localhost:5001/sdui which you may prefer

[yurivict]

This worked.

Thank you for your help!