The model copies (as much as it turns out) the AndroidArts game designer's style.
Unfortunately or fortunately the author Arne Niklas Jansson (AndroidArt) draws mostly people, machinery and electronics, because of this model can poorly draw landscapes, backgrounds without people...In the next versions I will try to fix this.
Pony Lora
Also you can familiarize yourself with the author's work here - Android Arts
Recommended parameters for more accurate style replication :
The keyword - "background" can create you just a white or any other single-color background.
Use - "indoors" or "outdoors".
Increasing the CFG Scale above the recommended amount can ruin your end result
Sampling method : Euler a
CFG Scale: ~4-5
Sampling steps: ~22
Quality tags : not required (optional)
Negative promt : easynegative (optional)
If you like what I do, you can support me here-- https://boosty.to/snzhkhd/donate
v3 - Additional pre-training. Quality improvement
v2 - Improved faces and overall style
You can reinforce merge with trigger word - androidarts
Description
Additional pre-training.
quality improvement
FAQ
Comments (12)
This is really well done. Thank you for sharing!
Big fan of the artist, very excited you made this. But most of what I generate ends up split into a 4 image grid or has tons of concept art style heads floating around, what am I doing wrong?
Whats the difference between V3 and V3 Alt
This is an excellent model. Could you give any details about your workflow? What base model, and software did you use? If you have any links to guides that you followed or found useful, this would be much appreciated.
I used the AnyLoraCheckpoint base model and Kohya Dreambooth. I don't have any guides, it's just normal model training and not a lot of experimentation.
What an amazing model!!! I have not seen any other model with such a cool style. In addition to it's great style, it is also great with anatomy, diversity and flexibility. Great job!