Which Flux model should I load in the desktop version?

[Hahaha-yap]

Good day, everyone. I’m seeking help from the community. Could you please tell me which version of Flux runs and works stably on a MacBook Pro M1 with 16GB RAM? Which versions of other models are you running without crashes? Thank you.

[LukeG89]

16GB is extremely low memory for Flux. But you could try the miracle with GGUF quantized models:
https://huggingface.co/calcuis/flux-dev-gguf

Here's the example workflow, you just need to install a custom node:
gguf_flux_dev_example.json

Here's the models you need:

• flux-dev-fp32-q2_k.gguf goes to ComfyUI/models/diffusion_models
• clip_l_fp32-f16.gguf and t5xxl_fp32-q2_k.gguf go to ComfyUI/models/text_encoders
• pig_flux_vae_fp32-f16.gguf goes to ComfyUI/models/vae

This should be the lighter setup for Flux Dev, let's see if you can make it work (success is not guaranteed)

[Hahaha-yap]

thanks a lot

[LukeG89]

@Hahaha-yap Is it working?

[Hahaha-yap]

I will try to download and check it today.

[Hahaha-yap]

@Hahaha-yap Is it working?

In short, it didn’t work for me with these files.

[LukeG89]

In short, it didn’t work for me with these files.

Ok, was kinda expected. That's a proof that there is no way to run Flux with 16GB

In fact, I think you can't even run SDXL with that memory (maybe a fp8 model)