No. All LoRA generations are private and never shared with anyone. CivitAi uploads are just free samples made by staff.
#1 Simplest LoRA trainer
no installation
no dataset
no config
5x-30x cheaper than a freelancer
Choose the model family
Choose type
References
Upload 1 to 3 images that represent the exact target result.
Uploading 0%
Trigger word
Pick the trigger word for the LoRA.
LoRA summary
Secure checkout powered by
Full dataset
Examples
Don’t worry about art style and backgrounds.
What matters
- 1 outfit only
- 1 character only
- No nudity (underwear is perfectly fine)
- Everything you want the model to learn must be clearly visible.
- Akilora does not train on these images directly. it builds its own understanding of the character, not the dataset. Show akilora what you want it to learn, no need for a massive dataset.
Full body information, last image is a face close-up to reinforce the face.
Full body information, first image is a close-up to reinforce the face.
Full body information, second image is a close-up to reinforce the face.
Full body information, but the face is not clearly visible, decreasing face accuracy. The shield obstructs the subject and may be learned. Low resolution of the last image is acceptable - Akilora will upscale it before training if needed.
Full body information, first image is a close-up to reinforce the face. Bonus points for showing the back, so the model can learn it too.
Full body information, first image is a close enough to learn the face properly.
Full body information, face close-up can be skipped since this character's face is pretty generic.
Outfit mismatch. The algorithm is designed to learn only 1 outfit and will only learn the face properly.
The algorithm does not understand chibi, miniature, pixelated, or otherwise simplified characters reliably. Convert them to a standard representation before uploading. ChatGPT is reasonably good at this:
Simplest LoRA trainer
-
1
Put in 1-3 images
-
2
Akilora trains it
-
3
Download your LoRA
It's that simple.
Try itPeople like it
How Akilora LoRAs perform on CivitAi:
Generations 328 800+ Downloads 53 200+ Upvotes 7 300+ Upvote rate 99.9%+
Examples
Follow us on Civitai for daily free models!
FAQ
Still have questions? Email us at hello@akilora.com and we'll help.
Please refer to the pricing page.
TLDR: Conservative, global average turnover time is 2 hours and 9 minutes.
It usually takes a bit more than a normal training tool (dataset building, preparation and preprocessing takes time) - the LoRA is usually ready to download within 2 hours. In rare cases, it can take more time.
No. Currently, only fictional, adult characters are allowed.
Akilora claims no ownership over images that were generated using our models, so you can do anything with them, including commercial use. Attribution is not required for images.
However, models generated using Akilora are another story. They are yours and you may share them, but Akilora is still the creator. You may not sell the models themselves, or claim that you hand-crafted them to your customers.
As a general principle, we allow all commercial usage of our models that does not break the law and does not involve directly reselling our models. If you are unsure about your use case, feel free to contact us at hello@akilora.com.
You can't. That's by design.
What values you put in depends on your dataset, and are pretty much unique every time.
Akilora's algorithm tries to calculate the best possible settings for any dataset. Messing with these values would break all of that, and at that point you are better off doing the whole thing yourself.
Akilora aims to be a simple tool for people who are not tech enthusiasts and would prefer to have someone else define things like epochs, mixed precision, or gradient checkpointing for them.
Try it
no installation
no dataset
no config
5x-30x cheaper than a freelancer
Choose the model family
Choose type
References
Upload 1 to 3 images that represent the exact target result.
Uploading 0%
Trigger word
Pick the trigger word for the LoRA.
LoRA summary
Secure checkout powered by
Full dataset
Examples
Don’t worry about art style and backgrounds.
What matters
- 1 outfit only
- 1 character only
- No nudity (underwear is perfectly fine)
- Everything you want the model to learn must be clearly visible.
- Akilora does not train on these images directly. it builds its own understanding of the character, not the dataset. Show akilora what you want it to learn, no need for a massive dataset.
Full body information, last image is a face close-up to reinforce the face.
Full body information, first image is a close-up to reinforce the face.
Full body information, second image is a close-up to reinforce the face.
Full body information, but the face is not clearly visible, decreasing face accuracy. The shield obstructs the subject and may be learned. Low resolution of the last image is acceptable - Akilora will upscale it before training if needed.
Full body information, first image is a close-up to reinforce the face. Bonus points for showing the back, so the model can learn it too.
Full body information, first image is a close enough to learn the face properly.
Full body information, face close-up can be skipped since this character's face is pretty generic.
Outfit mismatch. The algorithm is designed to learn only 1 outfit and will only learn the face properly.
The algorithm does not understand chibi, miniature, pixelated, or otherwise simplified characters reliably. Convert them to a standard representation before uploading. ChatGPT is reasonably good at this:
