Using instruct and mmproj as sdclip1 and sdclip2 loads model twice

[tuxArg]

Instructions for loading Qwen-Image-Edit say that we should use --sdclip1 and --sdclip2 parameters and point them to instruct and mmproj qwen2.5-vl gguf files.
But if we want to use qwen2.5-vl too as:
--model qwen2.5-vl-instruct.gguf --mmproj qwen2.5-vl-mmproj.gguf

Then the model is loaded twice (and uses 2x vram).

[LostRuins]

Yes unfortunately the image generation loads it very differently from text generation. It is not possible to share them both