Errors received and question on K2
Hi, thanks for the hard work and the FANTASTIC models.
I have a few questions:
1- What are the differences between the K2 and non-K2 model files?
2- Are they really SD3 models or SDXL? I ask because I get errors in my workflow. If I use the 15GB version it seems to work
Thank you again.
The k is just version place holder, so Original_Model then a update it become Original_Model_k2 and another update would be Original_Model_k3. I know most use a v1, v2, v3,
This model is an SDXL model with SD3 merged into it. EclecticEuphoria_Universal + SD3
The reason for k2 is that the SD3 model had the perturbed script ran against it and then merged with EclecticEuphoria_Universal where as the original model is the the sd3 model untouched.
As far as your Error, I have never seen an error message like that, not sure what to tell ya on that.
Could try to right click on the node and choose Reload Node or Fix node
You're awesome... thank you!!
Sorry, so one more question: Should I use it as SD3 or as SDXL? I thought the models were very different so I am confused how the model is merging SD3 and SDXL :(
I assume if it is SDXL none of the SD3 tooling / pipeline will work (16 channel VAE, SD3 LoRAs, Model Shift, SD3 latent, etc.?
Thanks again.
Use it as SDXL, but it won't work with Lora and such. I find it good for the initial image generation and then pass that image to another model to build on top of that starting image. It has VAE baked in, it's an advance block model merge. See the attached wf for an example - you can also reverse the process and use it as a refiner, but SD3 suppose to have good prompt adhesion so I put it first. I might be wrong on the LORA, seems it is, I don't use LORAs much
You're awesome ... many many thanks for all you do. ;)
Cheers!