Just a basic Krea 2 with Nvidia PiD workflow.
This workflow is preinstalled in my Qwen/ZiT/Boogu/Krea2 Runpod template
Note:
PiD only works at 1:1 ratio, so make sure to not change the AR from 1:1.
If you want to generate in a higher resolution than 1024x1024 to get more detail, you absolutely can, there's a subgraph that downscales the image to 1024x1024 before going through PiD.
Model downloads:
Krea-2 Turbo mxfp8 — 13.5 GB
Qwen3-VL 4B fp8 text encoder — 5.24 GB
Gemma 2 2B PixelDiT text encoder — 5.23 GB
Qwen Image PiD checkpoint — 2.72 GB
Qwen Image VAE — 0.25 GB
Description
FAQ
Comments (5)
Sixteen times the detail of Tittyout 4.
What's the purpose of Gemma 2 2B PixelDiT text encoder — 5.23 GB model? It doesn't look like it slots in anywhere in the workflow?
It's in the subgraph...
@bradcausa9919 Ah! My bad....I'm an old-school comfy UI user and I'm still getting used to sub-graphs. Thanks!
Now we need a workflow to batch-resize images to squares 😂