*NOTE: CosXL merged models only seem to be supported by ComfyUI at the moment.
There were several models that I really liked the results of, but I didn't want to have to keep swapping between them to get the results I was after, so I painstakingly went about merging them. This took many hours of merging and testing different models, in varying amounts, and comparing the output from many prompts until I was pleased with the final images.
I believe it is now ready! The example images were made using the default workflow in ComfyUI. There's no upscaling, face restoring or any other post-processing, and they should hopefully all have the workflow information embedded in each image.
I purposely haven't described the capabilities of this model, because it is extremely varied and very capable of MANY different types of styles. It has realistic skin, so can create gorgeous portraits, but can also do so much more!
I sincerely hope you like and enjoy this model, it's one I know I will be using a lot and probably for the majority of all my images, if not all of them. ;)
If you like my creations, then please consider supporting me with a coffee. Thank you!
Ko-Fi
Description
This is the previous version merged with CosXL to give a much richer range of colours, and true darkness. My "Go to" model for almost everything!
FAQ
Comments (13)
I'm getting garbled results on this latest version (2.5). Could you please verify if it's uploaded properly?
It says it is cosxl so may need a different sampler than usual?
Not heard any other reports of problems. VAE is baked in and should work fine with most samplers, I usually use DPM++SDE Karras with CFG between 2-5 and 30 steps.
same issue. using Forge.
Me too. I wonder what causes this.
I even re-installed from scratch, still same issue.
This does seem to be an issue with ForgeUI. I tried using the same model (which works in ComfyUI) in ForgeUI and I also get garbage output. Perhaps the UI requires changes to support this type of model...?
@galaxytimemachine I am using Automatic1111
Edit: indeed, it works flawlessly in ComyUI. Strange...
It seems that model (merged with CosXL) is currently not supported with Automatic1111 and Invoke that I'm using but works fine in ComfyUI. So, there is nothing wrong with your model's upload, it's just another type.
@robbedoes It's nothing wrong with the model file, it's just not yet supported on other UIs. I presume they can't use CosXL either.
Do I just use this CosXL model like any other SDXL model in Comfy? Or is there a special workflow for CosXL like Cascade has? I've never really created anything with CosXL. It sort of hit the scene and then vanished. 🤷🏼
Yes, use it like any other model. There are some samplers that work better than others, but experiment :)
This is a great model, one of my favorites. The main issue, is that it’s relatively hard to make it work! For example in Draw Things, it has to be imported with model objective v-prediction, but ALSO conditioning noise AND noise discretization DPM. Took some trial and error to figure out (plus a search for cosxl, that put me on the right path after finding the hugging face page) but it was worth it! :D
Thanks for the comment, and persisting with it! :)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.