pastel-mix

Update Logs:

[1/27/22]

I uploaded the model in CivitAI! -> https://civitai.com/models/5414/pastel-mix-stylized-anime-model I'd appreciate the ratings, thank you!

[2/2/22]

Uploaded a lora version.

<center><h1><b>Pastel Mix</b></h1></center> <p align="center">Welcome to Pastel Mix - a stylized latent diffusion model. This model is intended to produce high-quality, highly detailed anime style with just a few prompts.</p> <p align="center">This model is made with the thought of imitating pastel-like art and the potential of mixing LORAs into a model altogether to create a fantastic mix. Recipe for this mix could be found below. Like other anime-style Stable Diffusion models, it also supports danbooru tags to generate images. </p> <p align="center">e.g. <b>masterpiece, best quality, upper body, 1girl, looking at viewer, red hair, medium hair, purple eyes, demon horns, black coat, indoors, dimly lit</b></p> <p align="center"><img src="https://huggingface.co/andite/Pastel-Mix/resolve/main/example-images/grid-0020.png"> <img src="https://huggingface.co/andite/Pastel-Mix/resolve/main/example-images/grid-0018.png"></p>

---

How to download with Git

git lfs install
git clone https://huggingface.co/andite/pastel-mix

🧨 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 = "andite/pastel-mix"
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")

Gradio

We support a Gradio Web UI to run pastel-mix:

Examples

masterpiece, best quality, ultra-detailed, illustration, portrait, 1girl
Negative prompt: lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts))
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 448x640, Model hash: 7edc8e08, Model: pastelmix-fp32, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires resize: 960x1280, Hires steps: 20, Hires upscaler: Latent

masterpiece, best quality, ultra-detailed, illustration, portrait, hakurei reimu, 1girl, throne room, dimly lit
Negative prompt: lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts))
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 448x640, Model hash: 7edc8e08, Model: pastelmix-fp32, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires resize: 960x1280, Hires steps: 20, Hires upscaler: Latent

masterpiece, best quality, ultra-detailed, illustration, 1girl, witch hat, purple eyes, blonde hair, wielding a purple staff blasting purple energy, purple beam, purple effects, dragons, chaos
Negative prompt: lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts))
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 448x640, Model hash: 7edc8e08, Model: pastelmix-fp32, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires resize: 960x1280, Hires steps: 20, Hires upscaler: Latent

masterpiece, best quality, ultra-detailed, illustration, close-up, straight on, 1girl, black hair, yellow eyes, red roses, chains
Negative prompt: lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts))
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2203084815, Size: 640x448, Model hash: 7edc8e08, Model: pastelmix-fp32, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires resize: 1280x960, Hires steps: 20, Hires upscaler: Latent

masterpiece, best quality, ultra-detailed, illustration, close-up, straight on, face focus, 1girl, white hair, golden eyes, long hair, halo, angel wings, serene expression, looking at viewer
Negative prompt: lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts))
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 240742293, Size: 640x448, Model hash: 7edc8e08, Model: pastelmix-fp32, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires resize: 1280x960, Hires steps: 20, Hires upscaler: Latent

So what the hell is the 'better-vae' version?

I merged the pastel-waifu-diffusion.vae.pt inside the model so you don't have to set up the vae anymore.

life so much ez now since you don't have to download the vae and set it up right?

What is pastelmix-lora.safetensors?

It's a lora version which is made from extracting the loras from pastel-mix using a script that is similar to add-difference method.

https://github.com/bmaltais/kohya_ss/blob/master/train_network_README.md

Guide

For the settings or parameters, I recommend using these settings.

Sampler: DPM++ 2M Karras

Steps: 20

CFG Scale: 7

Hires. Fix: On

Upscaler: Latent (MUST!)

Hires Steps: 20

Denoising Strength: 0.

I prefer using 0.6 since it's the sweet spot of this model. If you can find a better setting for this model, then good for you lol.

Latent upscaler is the best setting for me since it retains or enhances the pastel style. Other upscalers like Lanczos or Anime6B tends to smoothen them out, removing the pastel-like brushwork.

Please use the VAE that I uploaded in this repository. It is from the Waifu Diffusion team. Credits to haru for letting me rename and upload it.

Tip (Optional)

Putting mksks style in the beginning of the prompt can further influence the pastel-like style and make the output better. It is optional though, so it's up to you. You don't really need it.

mksks style, masterpiece, best quality, upper body, 1girl, looking at viewer, red hair, medium hair, purple eyes, demon horns, black coat, indoors, dimly lit
Negative prompt: lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, low quality, normal quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts))
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 580841049, Size: 448x640, Model hash: 7edc8e08, Model: pastelmix-fp32, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Hires resize: 960x1280, Hires steps: 20, Hires upscaler: Latent

Recipe

Merging the models.

Merging the loras into the model.

License

This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage. The CreativeML OpenRAIL License specifies:

1. You can't use the model to deliberately produce nor share illegal or harmful outputs or content.
2. The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license
3. You may re-distribute the weights and use the model commercially and/or as a service. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully) Please read the full license here

---

Big Thanks to

The 東方Project AI community for their wonderful LORAs.

• Closertodeath for dpepmkmp model, and the loras: xlimo768, dpep 2 768
• dolphinz/sometimes#9353 for tabi artstyle Lora.
• SatyamSSJ10 for Jordan_3 Lora.
• randomaccessmemories#4004 for Magic Lora