All images out of the box, no postprocessing. Model
works best with high sampling levels. Larger resolutions
are also working good. If you have enough vram sometimes
no upscaler needed.
Recommendations (default use):
Optimum CFG-Scale: 7
Optimum Sampling Level: 150
Sampler Best Results:
Euler a
DPM++ 2S a Karras
DPM++ 3M SDE
DPM2 a Karras
Restart
The model is highly optimized on the settings
above, so you should leave out quality prompts
first, if not necessary.
The model is build of the latest trained models
and lora, thanks to all community members
for their work.
I hope you enjoy and share your images.
Ultrium is always a merge of the best and/or latest
avaiable models. So each Ultrium is different than
its predecessors. It's your decision if or which
Ultrium is best for you.
Ultrium 9.333 will be the last Ultrium for SDXL 1.0,
I will see for which AI I continue my work.
Description
Great Update, new merges, new embedded functionality, finetuned fingers
FAQ
Comments (11)
Dayum! All those prompts give exact what is advertised.
Thanks, that was my intention. Text is a little tricky and nude males are not working (the is no gut model to add for this functionality), there will be an update soon to improve this.
what are the main differences on the full vs pruned version?
I have not tested if the Quality is equal, but if "pruned" means, that very small values are cleaned, there must be some drawback. The full version is twice the data size, so this can be an issue on old machines, the rendering time is not so different. Loading time is no issue using SSD.
The way I understand it is that full is more for merging models and creating lora or embeddings.
Alright, this is the first time I've ever seen an XL model do NSFW better than 1.5 without using any LORAs at all. Just incredible!!! Now I'm convinced that XL will be able to do NSFW better than 1.5.
What render resolution do you use normally?
1024x1024?
Edit; I wanted to add that so far everything I've handed to this model has been rendered either incredibly well or at least is passing enough to be acceptable by most people's standards. I love it! It does run slower than other models, BUT, the end results are incredible.
Thanks for the praise. If you use the recommended samplers (see model text) you can use any resolution your vram is capable of. Ultrium is good in generating bigger images. The bigger the image the risk is increasing that things are doubling or welting weird. For example: If you use 2048x1024 instead of 1024x1024 there is a risk that content is created doubled. If you use 2048x2048 if could quadruple.
Resolutions that are smaller than 1024x2 are relative save.
Please Note:
It is also content dependend: Landsapes f.e. work great in most any resolution.
什么是vram???
Has anyone had any luck running v4 in DrawThings? I've tried both pruned and unpruned and they both give me an unfinished mess of noise.
[UPDATE] I had success manually installing the pruned version by copying the download link in the files popopen menu under the model details section. I suspect I was hitting memory cap when trying to cache the unpruned model.
Hello, drawn content and anime vanished partially in this version, I will try to bring them back. That is the difficulty - if you tune one part another may vanish. Stay tuned :) Feedback is always welcome.
TensorRT Engine Generation: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)" Clean AWESOME1111 WebUI 1.6 build also gives the same results across multiple resolution generations. Oh well..
[EDIT: Note that TensorRT is not compatible with SDXL models, but apparently it can with a rare select few. Wanted to check if it worked with this one.]
Works great and is slower, but results speak for themselves. I love this model very much ♥
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.