If you like what I do, leave me a ❤️ to receive notifications of future updates and follow me here on Civitai. Remember that it is a hobby for me and it costs a lot of time and everything is shared entirely for free, I appreciate that you make a review and share your creations on this wall.
Finally, a version of Consistency Corrector tailored for Illustrious-style models.
This LoRA was built specifically to enhance prompt adherence and internal consistency across compositions in Illustrious-based outputs — whether you're using the official release or any fork based.
🧠 What does it do?
As the name implies, it corrects typical issues in illustrious-based workflows:
Helps models better follow prompts
Reduces anatomical drift, pose desync, or detail inconsistency between similar frames
Improves repeatability in img2img, txt2img, and even partial inpainting chains
You can see a clear side-by-side example in the showcase images:
LoRA OFF (left) vs LoRA ON (right). Same prompt. Same sampler. Same seed. No negative prompt. Just consistently better.
🔹 Trained on Illustrious-compatible datasets (not generic SDXL)
🔹 It works perfectly with a strength 1, but you can adjust according to your needs.
🔹 Fully optimized for prompt-rich workflows with rich details, hands, symmetry, and pose logic
Small fixes and a new version of this LoRa.
It’s based on V1 but includes some very useful adjustments if the checkpoint we’re using has adherence issues. This LoRa fixes poor checkpoint adherence, especially with long prompts, whereas V1 only improves adherence and consistency for shorter prompts.
I hope you like it and show it some love!
I trained this LoRa specifically with the goal of automatically correcting small imperfections in generated images and thus avoiding the hassle of having to go to inpainting to try my luck. This tool has been ready for quite some time, but it took me a few months to do a lot of testing because, basically, it exceeded all my expectations, especially in terms of its very high compatibility with other image enhancement tools.
I highly recommend using it in conjunction with Midjourney Image Enhancer XL Extreme. Although it is not mandatory, since Consistency Corrector XL on its own is very powerful, the combination of both results in amazing images.
Recommended strength: 0.4
Used in conjunction with Midjourney Image Enhancer XL Extreme, the recommended strength is both 0.4
Helps models better follow prompts
Reduces anatomical drift, pose desync, or detail inconsistency between similar frames
Improves repeatability in img2img, txt2img, and even partial inpainting chains
