Ideogram4 v1.0
Just dropping a quick update! I just uploaded a new indovibe lora trained on Ideogram 4 using the same concept. Honestly, I noticed that Ideogram can capture details way better. But to be real with you, the likeness is just kinda average right now. I think I haven't figured out the optimal training parameters for this one yet. Plus, I had to use a smaller dataset this time around since writing all those captions took forever haha. if you're looking to generate NSFW content, the Krea 2 version is definitely still superior in terms of trainability. Anyway, feel free to play around with this version for some highly detailed generations and let me know what you think!
Krea2
v1.0
This is my very first LoRA trained using Krea.
The core concept remains exactly the same as my previous versions trained on Flux Klein 9B. However, I noticed that Krea's learning capability is significantly better and much more accurate than Flux Klein.
This Krea version has NSFW capabilities. I am quite satisfied with the results so far. However, since this is my first attempt with Krea, you might still notice some slight inconsistencies with the Indonesian facial features. Rest assured, I will continue to fine-tune and update this in the future.
v2.0
For this update, I've expanded the dataset, and I can confidently say that the overall results are noticeably better and more consistent now. That being said, to be completely real with you, I'm still not 100% satisfied with the current output. There's definitely still room for improvement, so rest assured I will keep fine-tuning and upgrading this in the future.
v3.0
Here is v3.0! For this update, I really wanted to push the boundaries of what this LoRA can do. I went all in and doubled the dataset size, making it 2x larger than before. I also put in the hours to completely clean up and improve the captioning. The combination of a much bigger dataset and better tagging means this version now understands a lot more concepts. It handles different outfits, environments, and specific prompts way better than the older versions while keeping the core likeness intact. The results are noticeably better, so feel free to test it out!
v4.0
Welcome to v4.0! This update is all about refinement and realism. I decided to completely overhaul and change the captioning for the dataset this time around. This massive change means the LoRA now understands a lot more concepts and adheres to your prompts way better. I also focused heavily on boosting the overall details and pushing the realism to the next level. The results are much sharper, more lifelike, and much more accurate to the core likeness. It's a solid step up from v3.0, so have fun testing it out!
Klein v6.5
This LoRA captures the authentic Indonesian beauty standard across three different age groups
How to Trigger (Important): You need to be specific with your prompts to get the exact age you want:
teenage indovibe southeast asian girlindovibe southeast asian womanmature indovibe southeast asian woman
Works best between 0.8 - 1.0, Great for boba shots 🧋, but please note it does not generate lower genitals.
Happy generating! Don't forget to share your results.
Description
Enhanced micro-details: Improved generation of overall textures, and intricate features for a much more realistic output.
Higher concept accuracy: The model now understands diverse concepts and subjects much more precisely, minimizing prompt bleeding.
Better trigger word separation: Clearer distinction and flexibility when utilizing various trigger words, allowing for more accurate generations.
Maintained the stability and anatomy improvements from the previous version.