There is a copy from this LoRa on a chinese site. They even copied my personal logo and name, so beware, where you download stuff.
I see several generations where you guys have <lora:text2:1> but it must be <lora:text:1> without the '2' or it does not work. A1111 has this behaviour.
And this is not a Pony LoRA, it does not work with Pony.
This innovative LoRa greatly enhances text generation endeavors. While not flawless, its success rate has notably increased.
Try incorporating the LoRa feature by employing 'text:"Hello"'. Don't forget to explore the provided examples for some additional inspiration or guidance.
The Strength can range from 0.5 to 1.4. Through my testing, I found optimal results with 'Euler a' and 'heun', yet feel free to experiment with other samplers of course.
I'm particularly eager for feedback this time around. Considering training a secondary model with additional imagery, i'm keen to understand any shortcomings or strengths before diving into it.
It is worth noting that this LoRa might struggle with lengthy prompts, such as extensive AI-generated essay prompts :D. Shorter prompts tend to yield better outcomes.
Licence:
Low Rank Adaptation models (LoRAs) and Checkpoints created by me (Smai-Lee).
As per Creative ML OpenRAIL-M license section III, derivative content(i.e. LoRA, Checkpoints, mixes and other derivative content) is free to modify license for further distribution.
In that case such is provided by licensing on each single model on Civitai.com. All models produced by me are prohibiting hosting, reposting, reuploading or otherwise utilisation of my models on other sites that provide generation service without a my explicit authorization. In the event of unauthorized commercial use, a fee will be charged. Also you are not allowed to use this LoRA in a merge and make it public anywhere.
Only civitai.com has the right to offer this LoRA for generation service.
Subsequent changes, linked to the evolution of new information technologies, have extended the scope of protection to photographic works, computer programs, databases and industrial design creations. Copyright is acquired automatically when a work is defined as an intellectual creation.
Description
FAQ
Comments (25)
Can it use for pony?
I don't use Pony, but feel free to try it :)
pony is really just XL so should work
@PlushBanshee it is trained differently I think, base SDXL loras don't typically work on pony
I highly doubt it works on Pony, none of the text ones do due to the base model
@Dever Now i am teased...
@SmaiLee Thanks for your lora.
No, after extensive testing and training a version for Pony, this Text LoRa does not work for Pony XL.
Unlike most other models, Pony has modified the underlying weight of SDXL base model so much that most LoRAs will produce rather distorted results.
For something as sensitive as text, it is not all that surprising that this LoRA does not work.
@NowhereManGo I meant, i trained some smaller versions especially for Pony, but it seems, there is not much text left inside the Pony modell a LoRa could work with. It is like SD1.5, can't make text reliable working for it.
@SmaiLee I see. That makes sense, given how much "retraining" went into pony.
Welcome to text hell
I've done some tests in CivitAI's Generator, using Pony Difffusion V6 XL... It don't work.
What if you make another version of this lora but for PonyXL?
If there is such demand, i can make some test trainings with a small dataset next week on Pony XL to examine the possibilities. But i won't promise anything at this point. Final training for this text LoRa took me approx 30h on a 4090 and i only have one ;)
First tests with 1k training images was very disappointing. Went trough 20 epoches with very common words, and only one or 2 hits. I ended up mixing in my current model and after 50/50 there were first signs of text restoring with one epoch. But Pony has lost its character already before that point. I also tried various combinations.
But as a result, i might upload Cine-Pony XL checkpoint :). Still needs a bit more testing.
Hello, I saw in a you tube video, he used 'Text' to get good texts. Greetings
Hi! When using your model, I get this message:
"Mature Content Restricted
A resource you selected does not allow the generation of Mature Content. If you attempt to generate mature content with this resource, the image will not be returned but you will be charged Buzz.
Text Aid SDXL - v1.0"
What is the reason for this? There is nothing about this in the description of the model.
I am fed up with all the big boobed p0rn gens on the page with incompatible Pony checkpoints, so i activated 'not for mature content'. Seems it does restrict the onsite generation as well. I am sorry for this, seems civitai has to implement a more detailed filter method.
Thank you for posting it, i keep that in mind next time.
@SmaiLee I click "remix" on my own image
https://civitai.com/images/17868302
and I see this warning.
It's funny)
@BadcucumbeR Strange, maybe post the issue on their Discord.
@SmaiLee Maybe you will do it, as a more reputable author?
any chance of a SDS 1.5 version?
Text works well. I found out that it works better with fewer words. Don't confuse the AI any more than it needs to be.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.