masterpiece, best quality, 1girl, green hair, sweater, looking at viewer, upper body, beanie, outdoors, watercolor, night, turtleneck
- HuggingFace Diffusers: https://huggingface.co/hakurei/waifu-diffusion
- Pickled Models: https://huggingface.co/hakurei/waifu-diffusion-v1-4
- Model Overview
- Training Process
- Prompting & Quality Augmenting
- License
- Sample Generations
- Team Members and Acknowledgements
The Waifu Diffusion 1.4 Anime model is a Stable Diffusion v2 model that has been finetuned from Stable Diffusion v2.1 Base.
The data used for finetuning Waifu Diffusion 1.4 Anime was 5,468,025 text-image samples that had been downloaded through an image board that provides high-quality tagging and original sources to the artworks themselves that are uploaded to the site.
Within the HuggingFace Waifu Diffusion 1.4 Repository are various models used within the production of the model:
- Waifu Diffusion 1.4 Anime Epoch 1: A test model made to properly ensure that the training setup works.
- Waifu Diffusion 1.4 Anime Inference Config: A file included to allow for inference with Automatic's WebUI and with the original Stable Diffusion codebase.
During dataset processing, the following quality modifiers were added to samples depending on the score of the post:
| Quality Modifier | Score Criterion |
|---|---|
| masterpiece | >150 |
| best quality | 100-150 |
| high quality | 75-100 |
| medium quality | 25-75 |
| low quality | -5-0 |
| worst quality | <-5 |
Additionally, certain metadata were also added in the form of rating modifiers.
| Rating Modifiers | Criterion |
|---|---|
| safe | Samples that are safe for work. |
| questionable | Samples that include risque visuals which are probably not safe for work. |
| nsfw | Sexually explicit or graphic material that you should not view while your boss is looking over your shoulder. |
| deleted | Samples that were marked for deletion. |
Based off of the above modifiers, an ideal negative prompt to guide the model towards high aesthetic generations would look like:
worst quality, low quality, medium quality, deleted, lowres, comic, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, jpeg artifacts, signature, watermark, username, blurry
And, the following should also be prepended to prompts to get high aesthetic results:
masterpiece, best quality, high quality, absurdres
masterpiece, best quality, 1girl, black eyes, black hair, black sweater, blue background, bob cut, closed mouth, glasses, medium hair, red-framed eyewear, simple background, solo, sweater, upper body, wide-eyed
masterpiece, best quality, 1girl, aqua eyes, baseball cap, blonde hair, closed mouth, earrings, green background, hat, hoop earrings, jewelry, looking at viewer, shirt, short hair, simple background, solo, upper body, yellow shirt
masterpiece, best quality, 1girl, black bra, black hair, black panties, blush, borrowed character, bra, breasts, cleavage, closed mouth, gradient hair, hair bun, heart, large breasts, lips, looking at viewer, multicolored hair, navel, panties, pointy ears, red hair, short hair, sweat, underwear
masterpiece, best quality, high quality, yakumo ran, touhou, 1girl, :d, animal ears, blonde hair, breasts, cowboy shot, extra ears, fox ears, fox shadow puppet, fox tail, head tilt, large breasts, looking at viewer, multiple tails, no headwear, short hair, simple background, smile, solo, tabard, tail, white background, yellow eyes
masterpiece, best quality, high quality, scenery, japanese shrine, no humans, absurdres
This project would not have been possible without the incredible work by StabilityAI. I would also like to personally thank everyone for their generous support in our Discord server! Thank you guys!
In order to reach us, you can join our Discord server.
The Waifu Diffusion 1.4 Weights have been released under the CreativeML Open RAIL-M License.
Why not add "lossy-lossless" to negative prompt?