CyberHarem/suzutsuki_kantaicollection

Lora of suzutsuki/涼月 (Kantai Collection)

What Is This?

This is the LoRA model of waifu suzutsuki/涼月 (Kantai Collection).

How Is It Trained?

• This model is trained with HCP-Diffusion.
• The auto-training framework is maintained by DeepGHS Team.
• The base model used for training is deepghs/animefull-latest.
• Dataset used for training is the stage3-p480-800 in CyberHarem/suzutsuki_kantaicollection, which contains 1224 images.
• Batch size is 4, resolution is 720x720, clustering into 5 buckets.
• Batch size for regularization dataset is 1, resolution is 720x720, clustering into 20 buckets.
• Trained for 10000 steps, 40 checkpoints were saved and evaluated.
• Trigger word is suzutsuki_kantaicollection.
• Pruned core tags for this waifu are long_hair, one_side_up, headband, grey_hair, breasts, hair_between_eyes, grey_eyes, white_hair, blue_eyes. You can add them to the prompt when some features of waifu (e.g. hair color) are not stable.

How to Use It?

If You Are Using A1111 WebUI v1.7+

Just use it like the classic LoRA. The LoRA we provided are bundled with the embedding file.

If You Are Using A1111 WebUI v1.6 or Lower

After downloading the pt and safetensors files for the specified step, you need to use them simultaneously. The pt file will be used as an embedding, while the safetensors file will be loaded for Lora.

For example, if you want to use the model from step 9500, you need to download 9500/suzutsuki_kantaicollection.pt as the embedding and 9500/suzutsuki_kantaicollection.safetensors for loading Lora. By using both files together, you can generate images for the desired characters.

Which Step Should I Use?

We selected 5 good steps for you to choose. The best one is step 9500.

1960 images (1.92 GiB) were generated for auto-testing.

The base model used for generating preview images is Meina/MeinaMix_V11.

Here are the preview of the recommended steps:

Step	Epoch	CCIP	AI Corrupt	Bikini Plus	Score	Download	pattern_0_0	pattern_0_1	pattern_1	pattern_2_0	pattern_2_1	pattern_3_0	pattern_3_1	pattern_3_2	pattern_4	pattern_5	pattern_6	pattern_7	pattern_8	pattern_9	portrait_0	portrait_1	portrait_2	full_body_0	full_body_1	profile_0	profile_1	free_0	free_1	shorts	maid_0	maid_1	miko	yukata	suit	china	bikini_0	bikini_1	bikini_2	sit	squat	kneel	jump	crossed_arms	angry	smile	cry	grin	n_lie_0	n_lie_1	n_stand_0	n_stand_1	n_stand_2	n_sex_0	n_sex_1
9500	32	0.961	0.980	0.816	0.847	Download
5750	19	0.956	0.987	0.822	0.842	Download
8250	27	0.948	0.961	0.830	0.823	Download
10000	33	0.944	0.974	0.821	0.793	Download
6500	22	0.941	0.967	0.827	0.789	Download

Anything Else?

Because the automation of LoRA training always annoys some people. So for the following groups, it is not recommended to use this model and we express regret:

1. Individuals who cannot tolerate any deviations from the original character design, even in the slightest detail.
2. Individuals who are facing the application scenarios with high demands for accuracy in recreating character outfits.
3. Individuals who cannot accept the potential randomness in AI-generated images based on the Stable Diffusion algorithm.
4. Individuals who are not comfortable with the fully automated process of training character models using LoRA, or those who believe that training character models must be done purely through manual operations to avoid disrespecting the characters.
5. Individuals who finds the generated image content offensive to their values.

All Steps

We uploaded the files in all steps. you can check the images, metrics and download them in the following links:

• Steps From 7750 to 10000
• Steps From 5250 to 7500
• Steps From 2750 to 5000
• Steps From 250 to 2500