Attention: See version details for a full list of trained styles, the list is too big for the example images alone. I recommend using the CFG rescale extension with this model for improved generation quality (Forge has this built-in with the LatentModifier extension).

V3.0 is now out. This is trained over illustrious and is not a big modification of the base model. Instead, V3.0 cools it down a little, makes it respect prompts a bit more and produce bad anatomy outputs less often. It has also improved on ~100 different styles, making them more reliable and easier to mix together. Check the version details for more info and the sample pictures for ideas for positive and negative prompt.

V2.0 is out. This includes a more extensive training run with close to 40000 images in the dataset. I recommend NOT using the Karras scheduler with your samplers (instead, you can use the regular version of said sampler. I use DPM++ 2M SDE). Use one / several of the 90 different style tags (listed in the "about this version tab on the right) for best results. Thanks to extra, nedzet and everyone else who put in the effort to test it. Special thanks goes to the folks at unstable diffusion for graciously lending compute to train this.

First version of Olympus in XL. This model is trained on top of animagine, using ~28000 curated images, trained for about 50 epochs on a cosine learning rate schedule. For best results, be thorough in your prompts. This model follows what you prompt very closely, which is good and bad. As a general rule, if you want something to appear, say it in the prompt and don't assume it will just happen (the classic example is nipples). See the example images for examples, they are straight text to image without any inpainting or adetailer. Also, make sure to use some of the 80 different styles available to the model. These were all trained in this version and don't require loras. That is the best way to achieve great gens with this model.

This is my first attempt at finetuning an XL model, and while the results are good, I expect this to improve significantly with next versions.

All images were generated without adetailer or inpainting of any kind, using only hires fix with 0.3-4 denoise.

License: Fair AI Public License