Gemini Fantasy: Bold Comic Ink (Krea2 Style LoRA)

This is a very special release for me, though this V1 comes with a few quirks! This style was carefully cultivated through a massive amount of iteration and prompt engineering between myself, Claude, and Gemini. My goal was to perfectly capture the incredible "Bold Ink Comic Book Fantasy" aesthetic that Nanobanana produces, but bring it natively into Krea 2 so we could have complete freedom over the output.

(Note: During the curation of this dataset, the AI and I actually ended up generating three distinct sub-styles. This one—Bold Comic Ink—is my absolute favorite. The other two—Painterly Splash Art and Soft Cartoon—are currently being prepped for future releases!)

⚠️ V1 Usage, Strengths, & Limitations

I'm releasing this as an Early Access V1 because further testing revealed a distinct dataset bias. The model was slightly overtrained on female subjects and as a result favors female subjects heavily with higher strengths.

• Recommended Strength: The sweet spot is between 1.0 and 3.0.

• The Quirks (Gender Bias): At lower strengths, the model follows prompts well while applying a nice bold-ink flair. However, as you push the strength past 2.0, the model will actively start trying to force your subjects to be female, regardless of your prompt.

• The Plan for V2: I will be re-training this style in the future with a re-balanced dataset to decouple the female subjects from the art style itself. Until then, V1 is highly usable as long as you keep the strength managed!

🛠️ Training Details & A Massive THANK YOU!

• Dataset: ~300 curated images

• Specs: 2000 steps, 20 Epochs, Rank/Dim/Alpha 16, LR 0.0001

• Trainer: Civitai On-Site Cloud Trainer

A HUGE thank you to this community! Because of the Buzz support I received on my previous Early Access models (which have all been made free, as promised!), I was able to reinvest that Buzz into Civitai's on-site trainer. My home GPU struggles with datasets over 100 images, so your support literally saved my hardware from melting and allowed me to train this larger dataset. Thank you!

☕ Support, Temporary BUZZ Access, & Commissions

As with my previous Krea 2 releases, this model is temporarily under paid BUZZ access. Once I recoup the Buzz spent on the cloud compute for this model, I will flip the switch and make it 100% free for everyone forever. Your early support directly funds the cloud training for my upcoming Krea 2 LoRAs (and the V2 retrains)!

I am also open for custom LoRA / LoKR training commissions (private use or expedited public release). Feel free to reach out via DM! (Note: I do not accept commissions to train on IRL persons).