Qwen Image simple GGUF workflow (16VRAM 32RAM)

How it's going gguffers, it's time to spit out some super simple and based txt2img workflow for goofy GGUF users like me who often struggle with installation or just don't care about advanced features and want just to touch a model.

Requirements:

• ComfyUI v0.3.49+

• ComfyUI Manager v3.35+

• 16GB of VRAM and 32GB of RAM (or less, if you will choose lower quants)

Installation:

• Download model files:

  • Main model (drop into ComfyUI\models\unet). Options: choose any from Q5_1, Q5_K_M, Q5_K_S or bigger. I prefer 5_1.

  • Text model (drop into ComfyUI\models\text_encoders). Options: UD_Q5_K_XL, Q5_K_M or bigger. I took Q6_K.

  • VAE

• Download and open this workflow in ComfyUI.

• Go to "Manager" - "Custom Nodes Manager" and install "ComfyUI-GGUF" v1.1.3 or above (older versions will blow an error "Unexpected text model architecture type"). Restart the ComfyUI.

Usage:

• Choose some resolution: any divided by 16 should work, but native options listed inside the workflow note (x1328 and its variations) are the best. Text is better render at native resolution.

• Choose number of steps and a sampler:

  • 15-20 steps for Euler beta/simple (1-4 cfg), Euler_cfg_pp simple (1cfg)

  • 8 steps with DDIM beta (1-2 cfg). If you're crazy bastard like me, you can even set 5 steps and try your luck (lightning lora is not required).

If you experience crashes at the VAE Decode node, try using lower quants (if Q5_1 crashes, check the Q5_K_S).

If you have any other errors, do a clean install of ComfyUI and Manager and repeat.

Things tested on ComfyUI Windows portable edition v0.3.49 with 32RAM and 16VRAM 5060Ti.