Paradox 2 empowers you to use unique content while offering even more control over your creations by generating exactly what you request.
Unlike models that rely on merging existing data, Paradox 2 produces purely original content.
A cutting-edge model which draws inspiration from an extensive dataset of far over 10.000 original images, trained at 180.000 steps vs the first 65.000 steps release.
We invite you to load the model onto your favorite Stable Diffusion application, give it a spin, and share your thoughts with us.
- Special thanks to the people who help me and who I care a lot about:
@MarkOREZ
@Kamikaze(Elon Musk)
@osiworx
@mix
@Thibaud
@piscabo
@NobodyButMeow
@masslevel
Description
172500 steps.
FAQ
Comments (15)
Been playing with this for some time now (few hours for 2 days) and I really like it, well done Thaevilone
Glad to hear you like it, thanks a lot for the support. :)
Since there are many models, keeping track of which settings is needed for models performs well is difficult. It is helpful to have recommendations for generating good images in the description. I say this because I run it offline and so I copy the description and save it on a text file that accompanies the model, and so as of the time I write the comment, the only suggestion is clipskip 2, and it is elsewhere than the description.
Hello there, there are plenty of examples you can look at. :)
Hope you have a good time.
In general, there are no fixed settings that work on all images. For most SDXL based model, including Paradox, a CFG of 3-7 works well most of the time. Number of step is 20-70 depending on the sampler. Generl rule of thumb is that lower CFG needs more steps. DPM++ 2M SDE Karras works well, so does DPM++ 2M Karras. If you want a "softer" look, try "Euler".
Also the use of LoRAs will affect your settings. Some LoRA needs low CFG, some needs high CFG. Some LoRA only work well at low number of steps, etc.
As always, run some tests and try to find the best setting for the types of images you like to generate and the kind of "look" you like.
As Thaevilone said, you can just download some sample images that you like, and copy the parameters and paste them into your text file to look at them offline later.
Played with it whole day but this model just keeps on giving. It's truly something special. One the best out there right now. Well done!
Thanks a lot for your support and review, glad to hear you like it. :)
Great model, I'm so impressed! Thank you
No problem, enjoy. Thanks a lot for your support and the kind comment. :)
When is Paradox 3 coming?
The dataset is ready but it's on halt becuase of the upcoming SD3 model. :)
Thank you. Oh, yeah. This model's cool. It really allows you to visualize things that not all models can. Very capable and creative. And because of SD 3, it's gonna be even more awesome
@vas_shishkin You are definitely this model's biggest fan 👍🙏.
The variety, quality and quantity of your images are all great!
Having used this model for 3 months now, I continue to admire it for unlocking its potential. It definitely has the ability to squeeze all the others. May Paradox continue to develop and improve. Regards to the creator.
Hey there, thanks a lot for your kind comment. Glad to hear you still like it. :)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.