☕ Support My Work: If you enjoy my local tools and models and want to support me with real cents rather than just platform Buzz, you can do so here: https://ko-fi.com/monsieurbaka. Your support directly speeds up new software updates and model releases!
Klein 4b
Requested by hawaiianpizz. Surprisingly, the model (4b) performs quite well. In my opinion, the images need some refinement. I generated a showcase at 1440 by 896 - which is supposedly a bit too high for this model. In any case, give it a try. Version 9b is still the best option if you’re critically short on VRAM -> download the Int4/int8 convrot version; it should work on most computers.
Anima
In my opinion, Anima is a model designed for drawing characters, which it does very well. Geometry and architecture aren’t its strong suits. However, this Lora creates an interesting effect in the background of images, so I decided to release it. For some 3rd party loras it may be used as a background enhancer. Consider Klein 9b and Krea 2 to be the main releases.
Krea 2
This release is specifically at the request of ElGordoAI.
Klein 9b
My personal favorite. The main release for this Lora model Klein 9b works absolutely wonderfully; it’s the only model that consistently renders English text and the expected geometry.
Visual Novel Backgrounds
A new LoRA specialized in generating visual novel style backgrounds.
This LoRA is designed exclusively for creating backgrounds (backdrops / sceneries) in classic visual novel aesthetic.
Key Trigger Words
VNsprite, visual novel style, background,
Recommended Settings
- Resolution: 1280×720, 1440×900
- Steps: 6-12 (good starting point)
- Sampler: res_multistep
Experimenting with different samplers and step counts can give varied and interesting results.
Style Description
The LoRA produces clean, atmospheric visual novel backgrounds with soft lighting, nice color harmony and a characteristic anime/VN look — perfect for school classrooms, serene nature scenes, forests, beaches, and everyday locations.
Quality & Usage Tips
- The better and more detailed your prompt — the better the result.
- If you struggle with writing prompts, I recommend using an LLM (like Grok or similar) and feeding it examples from the Showcase and Gallery.
Known limitations (Klein 9b):
Occasional issues with complex architecture, precise perspective, shadows and lighting in very complicated scenes. In such cases try different seeds or do minor fixes in Photoshop/Post-processing.
This LoRA works very well for img2img editing.
Example workflow:
Base generation:
VNsprite, visual novel style, background, classroom interior, [detailed description], bright midday sun high in the sky...
Edit (img2img):
Change to: VNsprite, visual novel style, background, [detailed description], night during heavy rain, dark stormy sky, heavy diagonal rain streaks, wet reflective surfaces, moody dramatic atmosphere...
If the model sometimes adds unwanted people/characters during editing — add no humans, no characters at the end of the prompt.
For maximum composition and geometry preservation, I recommend using this supporting LoRA together with this one.
You can also try combining it with character LoRA to create CG scenes, but results may be unpredictable with complex compositions.
Training Information
- Dataset: 95 images
- Main source: ~60 images from Clannad
- Nature & forests: ~20 images from Everlasting Summer
- Modern buildings: ~15 images from Love Money Rock'n'Roll
Description
FAQ
Comments (9)
i haven't tested this and i'm already in love with it
Can you take existing images and turn them into visual novel style?
Thanks
Yes, that's possible. You can use any freely written prompt that involves editing.
For example: “Change to: VNsprite, visual novel style, background.”
Ideally, you should at least briefly describe the scene so the model understands what it’s working with. But it works without that too. I’ll upload an example to the gallery; it’ll take a while to load. Civitai takes a long time to load new images.
Someday we'll have models that will also build a 3D representation like BERT, then creating consistency will no longer be stupid-work.
can you do this one for krea2?
Pretty sure I can. The result might differ from Klein's. I don't use Krea, but I do have the ability to train the LORA model on it.
Wait about ten hours. If Civitai won't lagging this evening, you'll get a notification.
Done
@MonsieurBaka i love you man