Type | |
Stats | 372 555 278 |
Reviews | (91) |
Published | Apr 15, 2025 |
Base Model | |
Training | Steps: 800 Epochs: 20 |
Usage Tips | Clip Skip: 1 Strength: 0.9 |
Trigger Words | crdhrs2, 3d, realistic, photorealistic |
Hash | AutoV2 54DFC12A0C |
Changelog 17/4/25:
Added a version specifically trained on Uncanny Valley in order to get a more consistent style throughout, only version v3.0 is trained on Uncanny, the other v2.0 and v1.0 are trained on Illustrious as a base and works with other checkpoints.
After wasting so many buzz on this, losing a whole first iteration dataset because Civit's website was too slow so it messed up the epochs, I actually went on a crusade in order to grab some free blue buzz in order to work on the dataset once again and re-do the training, I'm still not really sure about this Epoch, so I'm uploading to get feedback since I ran out of Buzz myself to test it out more.
I ran out of buzz in order to hiresfix the last two images, but they overall came out really good even in low resolution. I'll wait till I get more buzz to generate more images and promptly show off more results for you guys, but for now... I'm done with v2.0 as it feels and generates very stable and consistent results, so yeah.
Recommended:
Strength 0.9 | CFG 3.5 or below, this is still in testing, as I did get better results using 4.5 on a few checkpoints.
I had this little comparison chart made by hand, because I don't generate locally, so you can use this as a guide to whatever checkpoint you're used to using.
Sorry, the quality of the image went to shit; here's a link for the comparison:
Thank you again for using my LoRA, you can support me through either donating Buzz, or you can use one of the links provided beneath my profile.