IMPORTANT!!!!
the workflow is subgraphed in a few areas to minimize the size of the workflow. it just cleans up everything. the subgraphs are set so you can activate your models in the nodes without having to open the subgraphs.
PLEASE MAKE SURE YOU ACTIVATE YOUR MODELS OR THE WORKFLOW WILL NOT RUN!
UPDATE YOUR COMFYUI AND CUSTOM NODES.
to install the DyPE node, simply search 'DyPE' in the comfyui manager custom nodes.
The DyPE node upscales images up to 4K natively without the need for extra weight on the workflow. originally created for Flux, its since been adapted to z-image and qwen.
node causes some issues with artifacting on z-image, i suggest keeping the upscale at 2-3K. 4k works but its sketchy sometimes. LoRAs are also a little sketchy sometimes with it.
Description
z-image+DyPE. no additives.
FAQ
Comments (10)
KSampler
Given normalized_shape=[2560], expected input with shape [*, 2560], but got input of size[1, 63, 3584]
show this error to me
activate your models inside the nodes. click the drop downs and select YOUR models, the workflow searches for mine by default. you have to activate yours for the workflow to run
I keep getting: apply_dype_to_model.<locals>.dype_patchify_and_embed() got an unexpected keyword argument 'ref_latents
yeah, they havent updated the custom nodes since all the updates hit for z-image. its causing everyone errors. i havent found a fix yet but when i do ill be sure to post it in this workflow page AND under your comment for you.
@realrebelai Thanks mate
@EvilPanda no problem! They still havent updated it yet btw 🤷♂️ idk what their doing.
@realrebelai @EvilPanda There's a workaround for ref_latents error: https://github.com/wildminder/ComfyUI-DyPE/issues/32#issuecomment-3795824066
@xmsthc it doesnt work sadly, i tried this last month when the the request was put in. Nothing you input into the code of the node changes it.
@realrebelai I think you might forget to patch some changes to the code, it works for me😂.
@xmsthc no youll see, your going to run into resolution issues