Current version 0.15
Normal Strenght of 0.6-0.8 & Pony version beween 1 - 2
I personally love this style, it is something in between refined and unrefined,
This version has around 15% of the data i have, and i think it produces good enough results for a early access release.
Latest release: Wizardry for Illustrious
Version: Pre-Alpha 0.1
Type: LoRA (SDXL)
Wizardry for Illustrious is an early-stage LoRA trained on WAI-illustrious-SDXL v16. This is a foundational release intended as a baseline for future iterations. The current version reflects experimental keyword structuring and strength calibration, both of which will be revised in upcoming updates.
Current Status
This is a pre-alpha build. Prompt keywords are likely overextended due to long text conditioning during training, which may affect consistency and control. Strength scaling is not fully calibrated at the default 1.0 weight.
Future versions will:
Refine keyword structure and reduce over-conditioning
Improve strength responsiveness
Adjust training balance for better baseline usability
Recommended Settings
Suggested weight range: 1.25 – 2.75
1.0 weight: Underpowered / weak influence
4.0 weight: Likely to destabilize outputs
For best results, begin at 1.5–2.0 and adjust incrementally.
Base Model Compatibility
Primary generation base: WAI-illustrious-SDXL v16
Alternative base: WAI-illustrious-SDXL v14 (may produce a more artistic output; not fully tested)
No additional LoRAs were used during testing. Given Illustrious’ strong baseline capabilities, stacking should be viable, but compatibility has not yet been formally evaluated.
Notes
This release serves as a structural baseline for the Wizardry line. Expect substantial changes in future versions as prompt architecture and training refinement are improved.
Description
New version, trying something new
FAQ
Comments (9)
Update? Sign me in!
its a small one, added 40% more data, and it was trained in a different way, lets see if this works, if not il revert some changes.
If I could only access this to review it... I enjoyed the first version already.
But I really dislike the early access feature. By the time it's publicly released there are 30 other new things to check out :/
its mostly because im not sure if im done with it, and was testing it out, but its out in 24 hours (15 left).
@xRob thank you very much! Wouldn't more people with access be more helpful in finding out what's good and what's bad in the current version though? You could specify aspects you're especially uncertain about and could use more data for by asking for specifics you'd like tested and experimented with. Like how consistent the LoRA performs across all sorts of checkpoints, or specifically photorealistic ones or how well it's doing in terms of strength/weight adjustment in different settings.
Ask users to share keywords or similar artists known to sdxl that might emphasize the style and pick the ones that you find fitting to add them to a list of suggestions for future users.
What I'm getting at is just that all of this is experimental in all stages and everything is bound to work better or worse in some scenario or another. Of course I understand that you don't want to release something you're unsure about, but what's the difference between releasing it to a smaller user-base versus a larger one, especially in terms of getting feedback to improve?
Are there any advantages in early-access-locking I just don't know about? Is civitai rewarding that in any way to support creators for "enticing" users to become paid members?
This is not meant to sound as harsh as it reads. I'm honestly curious and support everything helping you as a creator as I greatly appreciate your efforts.
It's just that early-access feels like being gate-kept out of contributing to a project, which seems like a lose/lose for everyone involved. Am I missing something obvious there?
@redpinkretro I know what you are trying to say, I don't have a pro-sub myself, but 24-36 hours is negligible and if it helps the site get signups I see that as a good thing. i like the site and its features and they need to make money somehow.
Since it's not permanent and it's not a long time I don't find it that problematic at this moment.
@xRob So it's a way for creators to actually support the site more, I understand that. Sure 24-36 hours is not as bad as two weeks, which is definitely long enough to entirely forget about a LoRAs existence considering the rate of new developments and changes.
Are the models bumped up on the list on public release? I don't think they are, but I'm not sure. Would definitely help compensate for the "I forgot this even existed" effect when a model is inaccessible for a longer time.
@redpinkretro I am unsure, maybe a support case could answer that one. Regardless ive set mine to 24 hours in the future. Should be acceptable!
@redpinkretro This would be a cool feature cause I get so many notifications that the out of early access ones get lost in the mix.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.