Nix comic is a flexible comic base - with a core style mixture of 1990s Disney animation and 2000s Marvel Comics. It is tuned to be used at 0.7-0.85 strength on most prompts, giving headroom for mixing in other LoRA or style guidance. No keywords, just add, enable, and generate.
V2
Similar core data set and and same goal, but much more flexibility, stronger definition of style, and richer scenes!
Refined V2 Foundation: Upgraded dataset with richer scene depth, enhanced background detailing, and cleaner base line art.
Precision Weight Merging: Blended with re-ranked and stripped LoRA to refine: Krea2 Zass-style Animation (body shape & background depth), Friendly Sketch Krea2 (line details & highlights), Nix Toon (color, highlights, tonality) using low-rank SVD reconstruction.
High Prompt Adherence: Text encoders stripped on secondary models to ensure prompt-defined colors, text, and scene details render cleanly without style hijacking.
Stackable & Diverse: Designed to prevent same-face/same-body bias, respect anatomical diversity, and stack effortlessly with additional character or style LoRAs at 0.8 – 1.0 strength.
Overbaked Variant
Author's note: These are left up for historical purpose only - their output is poor (saturation/shadows/details are all a mess). They were an attempt to apply the style in some of the NSFW and artsy checkpoints - but really just don't do well anywhere.
Why not Nix Toon v2?
This started as Nix Toon v2, but the core style drifted quite heavily. This is what I intended Nix Toon to be - extensible, but a strong core style. While re-tuning that model, I worked to eliminate same-face and the unintentional loss of cultural identity. While it's not perfect, Nix Comic is much more open to more body shapes/sizes/colours.
What's different from other Comic/Toon LoRA?
Nix Comic has a strong visual identity, with ability to take direction from other LoRA and additional style prompting (without getting too crunchy, overbaked, or breaking style). It's also built for diversity, where possible.
Special callout to @ZairaSass - some of their LoRA stack was used for inspiration.
Description
This model is technically trained further than I normally would like. It may help those using it on heavily trained/merged checkpoints that don't favour illustration.
FAQ
Comments (3)
Stunning results! I'm glad that my style stack was helpful as you made this. I'm excited to play around with this 👏🏻🙌🏻👏🏻
Thanks for the note! Happy for any feedback you have. Appreciate your work and style!
Putting it here too: NSFW / Heavy Modified models may need the Overbaked version + illustration/cartoon/disney key words to tap into the styles.