Welcome to Hitokomoru Diffusion - a latent diffusion model that has been trained on Japanese Artist artwork, ヒトこもる/Hitokomoru. The current model has been fine-tuned with a learning rate of 2.0e-6 for 20000 training steps/80 Epochs on 255 images collected from Danbooru. The model is trained using NovelAI Aspect Ratio Bucketing Tool so that it can be trained at non-square resolutions. Like other anime-style Stable Diffusion models, it also supports Danbooru tags to generate images.
e.g. 1girl, white hair, golden eyes, beautiful eyes, detail, flower meadow, cumulonimbus clouds, lighting, detailed sky, garden
There is 4 variations of this model available so far:
hitokomoru-5000.ckpt for the checkpoint trained on 5,000 steps.hitokomoru-10000.ckpt for the checkpoint trained on 10,000 steps.hitokomoru-15000.ckpt for the checkpoint trained on 15,000 steps.hitokomoru-20000.ckpt for the checkpoint trained on 20,000 steps.
Dataset
You can find datasets used to train this model and the last-state folder for resume training here
Diffusers
This model can be used just like any other Stable Diffusion model. For more information, please have a look at the Stable Diffusion.
You can also export the model to ONNX, MPS and/or FLAX/JAX.
from diffusers import StableDiffusionPipeline
import torch
model_id = "Linaqruf/hitokomoru-diffusion"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
prompt = "hatsune_miku"
image = pipe(prompt).images[0]
image.save("./hatsune_miku.png")
Examples
Below are some examples of images generated using this model:
Using Hitokomoru-5000-pruned.ckpt
Anime Girl:
