Version AisakaPony_v4

I found that the stable diffusion could not display v3, I created a new one, and some improvements have been made

The recommended weight is 0.7 to 0.8.

Version AisakaPony_v3

Adding more training data, Improved the issue with the buttons.

This time makes me feel that it will be clearer than the last version.

More Kawashima Ami and Kushieda Minori !! But it's still difficult to get them in the same frame.

It seems I've spent too much time on her...

But at the same time, I've also learned a lot -_+

ps:Now it's harder to get buzz... I'm sad :(

===========================================================

Version AisakaPony_v2

Thanks a lot to @Stagnation for giving me a lot of buzz! This time around, I've got more chances to experiment and achieve better results :)

I've continued to update the training set and tried to include Kawashima Ami and Kushieda Minori(Minori's face can sometimes look quite comical@_@), so entering their names and characteristics can generate them. However, it's limited to single mode, and it's quite challenging to have all three of them appear in the same frame at the same time. Also, there's no control over height. (By the way, Takasu Ryuuji might show up too, but I didn't specifically tag him.)

This lora is primarily themed around Aisaka Taiga.

In fact, there are several good outcomes, and it's really hard for me to pick the best one. I've spent three all-nighters testing to see which result is the best. Making a choice is quite a headache. My 970M graphics card has been put through its paces .I think I can finally take a good rest now(with my PC). I sometimes wish the world would end soon, just so I could rush to the store and grab a computer with better performance:)

Of course, that's all just me joking around. I sincerely hope for world peace. :)

PS: I've noticed the issue with the buttons on the clothes, and I'm looking for ways to solve it.

Adding 'button' to the negative prompts can reduce the number of buttons somewhat.

==========================================================

Version_AisakaPonyC

There will be some changes to the prompt words this time.

I trained with CivitAI and improved most of the training set. But due to carelessness, the file turned into 218MB and I spent all my buzz.

Now I can only earn buzz by constantly liking others' posts...

The results this time are very suitable for generating NSFW content...-_+

A new version will be available soon. I still have some doubts about the tags and am trying to figure them out.

The recommended weight is 0.75 to 0.85.

==========================================================

Version_1.5

又经过了一些尝试，稍微改进了一下。虽然没有达到预期的效果，但这也花了我不少的时间去尝试与学习，我还是决定发布一个比较好的结果。

After some attempts and minor improvements, although it didn't quite meet the expected outcome, I've spent a considerable amount of time trying and learning. Nevertheless, I've decided to publish a result that I consider relatively good.

==========================================================

Pony version!!

最近找到了新的colab训练lora，尝试做了一个ponyXL版本，效果比起之前好了很多。虽然细节还是有些模糊，但出现的效果也挺不错。我的电脑无法运行webui的XL模型，我可怜的970m每天都在发出悲悯，图片全由comfyUI生成。相信我，comfyUI要有趣的多。如果无法出现红色的校服，就稍微加大校服的权重。

Recently, I found a new Colab for training Lora, and I tried making a PonyXL version. The results are much better than before. Although there are still some fuzzy details, the overall effect is quite good. Unfortunately, my computer can't run the XL model on webUI, and my poor 970m is constantly in distress. All the images are generated by ComfyUI. believe me, ComfyUI is much more fun. If the red school uniform doesn't appear, just slightly increase the weight of the school uniform.

===========================================================

SD 1.5!

这是来自龙与虎的逢坂大河，经过很多的尝试得到了一个相对满意的结果，2D模型相对友好，真实类的模型就是一个灾难。权重推荐0.65~0.8，推荐用上dynamic-thresholding插件。

关键词：aisaka taiga, 1girl, solo, skirt, long hair, blue skirt, school uniform, pleated skirt, black ribbon

(个人喜欢加上黑丝）

以及我需要建议来改进这些模型:)

This is from Toradora! The character is Aisaka Taiga. After many attempts, I got a relatively satisfactory result. For the 2D model is relatively friendly, but for the realistic model is a disaster. I recommend weights between 0.65 and 0.8.and use dynamic-thresholding.

prompt: aisaka taiga, 1girl, solo, skirt, long hair, blue skirt, school uniform, pleated skirt, black ribbon

And I need suggestions for improving these models :)

Search AI Models & Addons

very few results