Companion tool to download model files?

[IngwiePhoenix]

I am setting up my own server and will use a mix of cards - ROCm for LLM and Intel (SYCL or Vulkan, whichever I can get to work reliably) - and I am looking into preparing the required files.

Unfortunately, unlike LLMs, image generation consists of quite a lot more files (encoders, VAE, encoder, ...) and it can get rather complicated...at least, to me. So, this might just be a me-problem. :)

Is there a tool - integrated or external - that can aid in downloading the proper files required?

I want to use Qwen-Image 2.1 - non-turbo, because I heared this allows the moden to be more detailed in it's output...but I could be quite wrong here.

Thanks!

[benitoalba]

Hi! For non-turbo Qwen-Image 2.1, I'd use the Hugging Face CLI to grab the matching model, VAE and text encoder from the project guide. For text-to-image, run:

hf download Comfy-Org/Qwen-Image-2.1 \
  diffusion_models/qwen_image_2.1_int8_convrot.safetensors \
  vae/qwen_image_2.1_vae_bf16.safetensors --local-dir models/qwen21

hf download Qwen/Qwen3-VL-8B-Instruct-GGUF \
  Qwen3VL-8B-Instruct-Q4_K_M.gguf --local-dir models/qwen21/text_encoders

Add --dry-run first to preview the files and download size. The guide provides the compatible file list; hf downloads it.

For image editing, also grab mmproj-Qwen3VL-8B-Instruct-F16.gguf from the second repo and pass it with --llm_vision. The guide has the launch commands too.

[IngwiePhoenix]

This will do!

Kinda wish we had something similiar to llama.cpp's -hf and -hfd and friends. Oh well, this'll work too. :)