Qwen Image, quantized | Invoke | Page 1

brisk dome May 1, 2026, 4:54 AM

#

I see that the Qwen Image PR (#9000!) said to install the full Qwen Image model to get image-encoder and text-encoder, even if you're using the quantized version of the transformer.

Anyone since figured out a way around that? I ain't got the extra 40 GB to spare.

I see city96/Qwen-Image-gguf links to some files under Comfy-Org/Qwen-Image_ComfyUI.

The text_encoders/…_fp8_scaled.safetensors file did import as a qwen3_encoder checkpoint, but the vae/….safetensors file identified as for the anima base model and I haven't found a way in the UI to convince it otherwise.

woeful badge May 1, 2026, 10:38 AM

#

There is no Support for Checkpoint type vae right now.

brisk dome May 9, 2026, 1:13 AM

#

Thank you for getting this in to rc2!

#

I'm impressed by how well this model manages to do even when quantized and using the 8-step turbo.

brisk dome May 9, 2026, 2:38 AM

#

Hmm. Haven't figured out the conditions yet, but I've gotten hit by the system OOM-Killer twice now when switching models from Quen Image Edit 2511 to Quen Image 2512.

#Qwen Image, quantized