AstraliteHeart commited on
Commit
02abf56
1 Parent(s): 8635c36

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +24 -8
README.md CHANGED
@@ -9,19 +9,20 @@ inference: false
9
  thumbnail: "https://cdn.discordapp.com/attachments/1020199895694589953/1020200601780494386/000001.553325548.png"
10
  ---
11
 
12
- ** WIP **
13
- ** THIS IS A PLACEHOLDER, RELEASE ON FEB21 **
14
-
15
  # pony-diffusion-v3 - "kept you waiting huh" edition
16
 
17
  pony-diffusion is a latent text-to-image diffusion model that has been conditioned on high-quality pony, furry and other non photorealistic SFW and NSFW images through fine-tuning.
18
 
19
- WARNING: This model is capable of producing NSFW content so it's recommended to use 'safe' tag in prompt in combination with negative prompt for image features you may want to suppress.
 
 
20
 
21
- With special thanks to [Waifu-Diffusion](https://huggingface.co/hakurei/waifu-diffusion) for providing finetuning expertise.
22
 
23
  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1mSharzQFPD4mIUxNwBG7LCLAzGUS8Bmy?usp=sharing)
24
 
 
 
25
  [![Join Discord Server to try next generation models](https://discordapp.com/api/guilds/670866322619498507/widget.png?style=banner2)](https://discord.gg/pYsdjMfu3q)
26
 
27
  <img src=https://imagedelivery.net/_wFNZAzgWNWPmneM1cyjcw/artifact/449d42a8-28c5-44da-afd7-28d7e29a264c/public width=25% height=25%>
@@ -32,8 +33,9 @@ With special thanks to [Waifu-Diffusion](https://huggingface.co/hakurei/waifu-di
32
  <img src=https://imagedelivery.net/_wFNZAzgWNWPmneM1cyjcw/artifact/f70fde75-f7ba-4c78-8229-3e39314aaf43/public width=25% height=25%>
33
  <img src=https://imagedelivery.net/_wFNZAzgWNWPmneM1cyjcw/artifact/422c26b6-38fa-4e0b-9c33-cffe4dc6f879/public width=25% height=25%>
34
 
 
35
 
36
- [Original PyTorch Model Download Link](https://mega.nz/file/Va0Q0B4L#QAkbI2v0CnPkjMkK9IIJb2RZTegooQ8s6EpSm1S4CDk)
37
 
38
  ## Model Description
39
 
@@ -41,6 +43,21 @@ The model originally used for fine-tuning is [Stable Diffusion V1-5](https://hug
41
 
42
  This particular checkpoint has been fine-tuned with a learning rate of 5.0e-6 for 20 epochs on approximately 1.7M pony, furry and other cartoon text-image pairs (using metadata from derpibooru, e621 and danbooru).
43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
  ## License
45
 
46
  This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
@@ -61,7 +78,7 @@ This model can be used for entertainment purposes and as a generative art assist
61
  import torch
62
  from torch import autocast
63
  from diffusers import StableDiffusionPipeline, DDIMScheduler
64
- model_id = "AstraliteHeart/pony-diffusion-v2"
65
  device = "cuda"
66
  pipe = StableDiffusionPipeline.from_pretrained(
67
  model_id,
@@ -88,6 +105,5 @@ image.save("cute_poner.png")
88
  This project would not have been possible without the incredible work by the [CompVis Researchers](https://ommer-lab.com/).
89
 
90
  - [Waifu-Diffusion for helping with finetuning and providing starting checkpoint](https://huggingface.co/hakurei/waifu-diffusion)
91
- - [Novel AI for providing compute](https://novelai.net/)
92
 
93
  In order to reach us, you can join our [Discord server](https://discord.gg/WG78ZbSB).
 
9
  thumbnail: "https://cdn.discordapp.com/attachments/1020199895694589953/1020200601780494386/000001.553325548.png"
10
  ---
11
 
 
 
 
12
  # pony-diffusion-v3 - "kept you waiting huh" edition
13
 
14
  pony-diffusion is a latent text-to-image diffusion model that has been conditioned on high-quality pony, furry and other non photorealistic SFW and NSFW images through fine-tuning.
15
 
16
+ **WARNING:** This model is capable of producing NSFW content so it's recommended to use 'safe' tag in prompt in combination with negative prompt for image features you may want to suppress (i.e. nudity).
17
+
18
+ **Despite its name, this model is capable of producing wide range of furry and cartoon images as side effect of improving data diversity (with exception of anime stlyes, for which Waifu Diffusion is much stronger choice).**
19
 
20
+ Special thanks to [Waifu-Diffusion](https://huggingface.co/hakurei/waifu-diffusion) for providing finetuning expertise and advising through the process, without their help this project would not exist.
21
 
22
  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1mSharzQFPD4mIUxNwBG7LCLAzGUS8Bmy?usp=sharing)
23
 
24
+ **Please join PurpleSmartAI Discord to use this model with our free SD bot and get early access to pony-diffusion-v4.**
25
+
26
  [![Join Discord Server to try next generation models](https://discordapp.com/api/guilds/670866322619498507/widget.png?style=banner2)](https://discord.gg/pYsdjMfu3q)
27
 
28
  <img src=https://imagedelivery.net/_wFNZAzgWNWPmneM1cyjcw/artifact/449d42a8-28c5-44da-afd7-28d7e29a264c/public width=25% height=25%>
 
33
  <img src=https://imagedelivery.net/_wFNZAzgWNWPmneM1cyjcw/artifact/f70fde75-f7ba-4c78-8229-3e39314aaf43/public width=25% height=25%>
34
  <img src=https://imagedelivery.net/_wFNZAzgWNWPmneM1cyjcw/artifact/422c26b6-38fa-4e0b-9c33-cffe4dc6f879/public width=25% height=25%>
35
 
36
+ You can see more samples at [PurpleSmartAI](https://purplesmart.ai/collection/top?nsfw=0&page=1&gen_type=txt2img&model=5&order=created_desc)
37
 
38
+ [PyTorch Model(Use this with Automatic1111 or other SD UIs)](https://mega.nz/file/9S0iSLpC#NqTvjJIzKoT2ScP9F5tFj7J3p5DhpLbIKiR70hMXqQ4)
39
 
40
  ## Model Description
41
 
 
43
 
44
  This particular checkpoint has been fine-tuned with a learning rate of 5.0e-6 for 20 epochs on approximately 1.7M pony, furry and other cartoon text-image pairs (using metadata from derpibooru, e621 and danbooru).
45
 
46
+ ## Improvements over previous models
47
+
48
+ ### Better disentanglement of tag based prompts
49
+ Aka ["using Hidden States of CLIP’s Penultimate Layer"](https://blog.novelai.net/novelai-improvements-on-stable-diffusion-e10d38db82ac#:~:text=Using%20Hidden%20States%20of%20CLIP%E2%80%99s%20Penultimate%20Layer), a technique adopted by SD2 which should lead to generally higher quality and more tag driven outputs.
50
+ In our experiments using penultimate CLIP was not always the best choice so trying both CLIP skip of 1 and 2 is recommended.
51
+
52
+ ### Better support for non square images and increase of default resolution to 768px
53
+ This should allow you to generate full body image with resolution 512x768px without triggering "double head" glitches.
54
+
55
+ ### Removed SFM/3D bias
56
+ V2 model demostrated bias toward 3d/sfw visual styles so special care has been applied to restrict exposure to 3d/sfm images during training.
57
+
58
+ ### Improver data diversity
59
+ We ran multiple experiments finetuning models on 300k pony only and 600k pony only datasets, the resulting models demostrated worse quality for pony specific data. We concluded that despite more complicated prompting (and lack of pony bias by default), inclusion of large amount of non pony specific highly ranked non photorealistic images generally has strong positive efect.
60
+
61
  ## License
62
 
63
  This model is open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage.
 
78
  import torch
79
  from torch import autocast
80
  from diffusers import StableDiffusionPipeline, DDIMScheduler
81
+ model_id = "AstraliteHeart/pony-diffusion-v3"
82
  device = "cuda"
83
  pipe = StableDiffusionPipeline.from_pretrained(
84
  model_id,
 
105
  This project would not have been possible without the incredible work by the [CompVis Researchers](https://ommer-lab.com/).
106
 
107
  - [Waifu-Diffusion for helping with finetuning and providing starting checkpoint](https://huggingface.co/hakurei/waifu-diffusion)
 
108
 
109
  In order to reach us, you can join our [Discord server](https://discord.gg/WG78ZbSB).