I am getting confused : sdclip vs. sdllm for image gen feature

[alex-ie]

It got me confused initially using kcpp for image gen to put LLM link into clip1 field in launcher, but I got used to. Now with 1.123 announced, new template has: https://huggingface.co/koboldcpp/kcppt/blob/main/Qwen-Image-2.1.kcppt:

"sdllm": "https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct-GGUF/resolve/main/Qwen3VL-8B-Instruct-Q4_K_M.gguf"
("sdclip1": "https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct-GGUF/resolve/main/mmproj-Qwen3VL-8B-Instruct-Q8_0.gguf"),

whereas several months old template:
https://huggingface.co/koboldcpp/kcppt/blob/main/Krea2Turbo.kcppt:

"sdclip1": "https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct-GGUF/resolve/main/Qwen3VL-4B-Instruct-Q4_K_M.gguf"

Does kcpp support giving link to LLM in both sdllm and sdclip1 fields now? Or sdclip1 usage for LLM is deprecated? Or else? TIA

[wbruna]

Both, actually. You should use sdllm from now on, but sdclip1 keeps being supported for compatibility.

[alex-ie]

A bit tangential question: why does QI 2.1 template contain mmproj file? I have not seen mmproj on official Qwen Image page, i.e. https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo/tree/main.

[wbruna]

why does QI 2.1 template contain mmproj file?

It's for edit mode.

I have not seen mmproj on official Qwen Image page

For gguf ffiles, the llama.cpp convention is to keep mmproj weights separated from the main LLM model.