This is a "from above" slider LoRA. I have used a selfmade tool to strip away unwanted data on layers that didnt affect the "from above" effect, and this have made this LoRA very clean and robust.
You don't need to prompt "from above" for it to work, but adding that kind of phrasing will reinforce the effect. When paired with character LoRAs or other heavy modifiers, you'll likely need to push the strength higher to get the angle to come through.
The model breaks down quickly once you cross its limits, and where those limits sit depends on your prompt and what else you have loaded.
In a clean prompt with nothing else fighting it, the sweet spot is around 4 to 8.
If your prompt already leans into a "from above" angle, that same 4 to 8 range will take it extremely high.
Stacked with heavy LoRAs, you may need to push it up to 8 or 10 before it starts hitting the limit. For best results use a prompt that leans into "from above" and use around 5-6
All my work is 100% free. The only thing that really matters to me is people leaving a review, good or bad. I will NEVER remove or block anyone for leaving a bad review, so don't be afraid to give me criticism.
I'm stating this because when a creator blocks someone for leaving a bad review, that person can no longer see any of the creator's models, essentially muting their critics. This is something I'm highly critical of, and I believe it's a fundamental quality problem on this site.
Description
FAQ
Comments (8)
I tested your 19 new LORAs, and apart from the fact that you removed the old angle sliders, the others worked quite well. (I noticed that your LORAs have very little impact on the faces of the original ZIT celebrity characters, which is fantastic. How did you do that? Most LORAs will directly and subtly adjust the facial features as the intensity increases. Did you crop the character's head during training?)
Thank you!
@wyxzddsjj919 When training a slider lora you dont retrain the model like a normal lora, you just multiply different parameters of the exiting model. Its almost like a prompt in slider-form.
The only downsite it that there is impossible to make loras of things the model dont have in its dataset already.
For this lora it just uses the paramters that triggers a image from a higher position and makes it much stronger then whats possible with prompts alone. But its the same "under the hood" as a prompt that say "from above".
I use ai toolkit. But what makes my loras so good compare to a lot of others is that i have a perfected the prompting needed for training, and some settings in ai toolkit ill keep for myself :P
@Loraholic 明白了 感谢大佬无私的分享~
On naming conventions, I rename all my loras anyway, and keep local notes on how to use them. For my part, I'd benefit most from your description on the page of how to use it and its particulars being well-written and concise. I copy-paste it into my notes, so if it's well-written, I don't have to rewrite it.
Thanks for all your LoRAs, btw - I've gotten a lot of use out of them.
We really need this one for Krea2 😊
Thank you for all your hard work, but are we really not going to see the lora of Krea 2?
No, sorry. Not in the near future