README.md

---
license: mit
datasets:
- hearmeneigh/e621-rising-v3-curated
- hearmeneigh/e621-rising-v3-finetuner
library_name: diffusers
pipeline_tag: text-to-image
tags:
- anthro
- furry
- e621
- nsfw
- booru
- imagebooru
- imageboard
- gelbooru
- danbooru
- rule34
- not-for-all-audiences
---

<div style='background: #ffeef1; border: 1px solid #fd91a4; padding:1em; border-radius:3px; margin-bottom:2em;'>
<h3 style='margin:0'>NSFW</h3>
<p style='margin:0'>This model  is not suitable for use by minors. The model can and will produce X-rated/NFSW content.</p>
</div>


<div style='background: #eefaff; border: 1px solid #91cefd; padding:1em; border-radius:3px; margin-bottom:2em;'>
<h1 style='margin:0'>Quickstart</h1>
<div style='margin:0; margin-top: 1em; margin-left:1em;'>
  <h3 style='margin:0'>Downloads</h3>
    <div style="margin-left: 1em;">
      <a href="https://huggingface.co/hearmeneigh/e621-rising-v3-safetensors/resolve/main/e621-rising-v3-epoch-34.fp16.safetensors" style="text-decoration: none !important;">
      ⤓ 
      </a>
      <a href="https://huggingface.co/hearmeneigh/e621-rising-v3-safetensors/resolve/main/e621-rising-v3-epoch-34.fp16.safetensors" style="text-decoration: underline">Checkpoint</a>
      <small style='padding-left: 0.5em'>
        (<a href="https://huggingface.co/hearmeneigh/e621-rising-v3-safetensors/resolve/main/e621-rising-v3-epoch-34.safetensors">fp32</a> | 
        <a href="https://huggingface.co/hearmeneigh/e621-rising-v3-safetensors/resolve/main/e621-rising-v3-epoch-34.fp16.safetensors">fp16</a> | 
        <a href="https://huggingface.co/hearmeneigh/e621-rising-v3-safetensors/resolve/main/e621-rising-v3-epoch-34.bf16.safetensors">bf16</a>)
      </small>
      <br />
      <a href="https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/resolve/main/rising-v3.csv" style="text-decoration: none !important">
        ⤓
      </a>
      <a href="https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/resolve/main/rising-v3.csv">Tag Autocomplete CSV</a>
    </div>
  <h3 style='margin:0; margin-top: 1em;'>Reference</h3>
    <div style='margin-left: 1em;'>
      • <a href="https://huggingface.co/hearmeneigh/e621-rising-v3/blob/main/INSTALL.md">Installation instructions</a>
      <br />
      • <a href="https://huggingface.co/hearmeneigh/e621-rising-v3/blob/main/CHANGELOG.md">What's new in v3?</a>
      <br />
      • <a href="https://huggingface.co/hearmeneigh/e621-rising-v3/blob/main/PROMPT-EXAMPLES.md" style='display: inline-block; margin-top:0.5em'>Prompt examples</a>
      <br />
      • <a href="https://huggingface.co/hearmeneigh/e621-rising-v3/blob/main/PROMPT-GUIDE.md">Prompt guide</a>
      <br />
      • <a href="https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/blob/main/tag-counts.by-name.json" style='display: inline-block; margin-top:0.5em'>Tag list</a>
      <br />
      • <a href="https://huggingface.co/hearmeneigh/e621-rising-v3/blob/main/AUTOCOMPLETE.md">Tag autocomplete guide</a>
    </div>
</div>
</div>

# E621 Rising V3 (SDXL)
* Furry / anthro base model trained with images (mainly) from [E621](https://e621.net) 
* Guaranteed **NSFW** or your money back
* Stable Diffusion XL 1.0 model:
  * `1024x1024px`
  * Trained with 11 epochs of 280,000 images each
  * Finetuned with 23 epochs of 40,000 images each
* Compatible with [🤗 `diffusers`](https://huggingface.co/docs/diffusers/index)
* Compatible with [`stable-diffusion-webui`](https://github.com/AUTOMATIC1111/stable-diffusion-webui)
* Fully open source crawl, dataset, curation, and training process:
  * Use these tools to train your own version with your own dataset!
  * [Configuration](https://github.com/hearmeneigh/e621-rising-configs)
  * [Toolchain](https://github.com/hearmeneigh/dataset-rising)
  * [Dataset](https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-curated)


## Examples
For more examples, [continue here](PROMPT-EXAMPLES.md).

<img src='https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/resolve/main/examples/singles/03-birb.jpg' alt='Birb' style='max-width:512px;' />
<img src='https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/resolve/main/examples/singles/02-royal-stallion.jpg' alt='Royal Stallion' style='max-width:512px;' />
<img src='https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/resolve/main/examples/singles/06-assassin.jpg' alt='Assassin' style='max-width:512px;' />
<img src='https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-preliminary-data/resolve/main/examples/singles/05-neekedneko.jpg' alt='Neeked Neko' style='max-width:512px;' />

For more examples, [continue here](PROMPT-EXAMPLES.md).

## Training Procedure

[Training legend](TBD)

* 160 images per batch (epoch variant)
* `1024x1024px` image size
* Adam optimizer
  * Beta1 = `0.9`
  * Beta2 = `0.999`
  * Weight decay = `1e-2`
  * Epsilon = `1e-08`
* Constant learning rate `4e-6`
* `fp16` mixed precision
* SNR gamma set to `5.0`
* Noise offset set to `0.07`
* `cosine_with_restarts` scheduler
* 11 epochs of [V3 curated dataset](https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-curated) samples resized to `< 1024x1024px` (maintain aspect ratio)
* 16 epochs of [V3 finetuner dataset](https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-finetuner) samples resized to `< 1024x1024px` (maintain aspect ratio)
* 6 epochs of [V3 finetuner dataset](https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-finetuner) samples resized to `< 1024x1024px` (maintain aspect ratio, randomly drop 70% of tags)
* 1 epoch of [V3 finetuner dataset](https://huggingface.co/datasets/hearmeneigh/e621-rising-v3-finetuner) samples resized to `< 1024x1024px` (maintain aspect ratio, randomly drop 50% of tags) and learning rate set to `4e-5`
* Tags for each sample are shuffled for each epoch