File size: 2,995 Bytes
dbaac11 c12a12b 83b8964 784c0a6 c0c00f1 784c0a6 dbaac11 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 |
---
tags:
- InstantID
- stable-diffusion
- diffusers
- text-to-image
- image-to-image
license: apache-2.0
---
# InstantID for StableDiffusion 1.5
This is unofficial implementation InstantID for StableDiffusion 1.5.
SD15 has a lot of finetuned models. So you can use all of this models with combination of instantid components to get awesome results.
<a href="https://instantid.github.io">Official InstantID</a> works only with SDXL and contains code only for inference.
But this repository contains Training and Inference code.
Training process was used only 10M images from <a href="https://github.com/FacePerceiver/LAION-Face">LAION-FACE 50M</a> dataset (Original InstantID used 50M Laion-face + 10M custom images).
Feel free to adapt it for your personal purposes. I will be glad if somebody find it usefull.
## How to use:
See <a href="https://github.com/TheDenk/InstantID-SD1.5.git">Github instruction</a>.
## Examples
Examples with <a href="https://civitai.com/models/132632/epicphotogasm">epiCPhotoGasm model</a> + styles from original InstantID.
<p>
<img src="./examples/demo/epicphoto.png" width="600" height="480" title="epicphoto"/>
</p>
Examples with <a href="https://civitai.com/models/65203/disney-pixar-cartoon-type-a">Disney Pixar Cartoon Type A model</a> + styles from original InstantID.
<p>
<img src="./examples/demo/disneypixart.png" width="600" height="480" title="disneypixart"/>
</p>
InstantID SD1.5 components are not compatible with InstantID SDXL. In this work model has been trained with additional facial keypoints information.
Keypoints visualization:
<p>
<img src="./examples/demo/kp_example.png" width="480" height="360" title="kp_example"/>
</p>
It is also possible transfer different keypoints from other images.
<p>
<img src="./examples/demo/pose_example.png" width="960" height="360" title="pose_example"/>
</p>
Links:
- <a href="https://github.com/TheDenk/InstantID-SD1.5">Code training/inference (gradio, jupyter notebooks, .py files)</a>
- <a href="https://huggingface.co/TheDenk/InstantID-SD1.5">Checkpoints (Controlnet, resampler, ip-adapter)</a>
- <a href="https://github.com/deepinsight/insightface/issues/1896#issuecomment-1023867304">Insightface models for keypoints (antilopev2 and others)</a>
## More examples
Using only models without special style prompts.
<p>
<img src="./examples/demo/rock.png" width="600" height="420" title="rock_example"/>
</p>
<p>
<img src="./examples/demo/mj.png" width="600" height="480" title="mj_example"/>
</p>
Examples with <a href="https://civitai.com/models/24387/aniflatmix-anime-flat-color-style-mix">Aniflatmix model</a> + styles from original InstantID.
<p>
<img src="./examples/demo/aniflat.png" width="600" height="480" title="disneypixart"/>
</p>
## Acknowledgements
- [InstantID](https://github.com/InstantID/InstantID) and InstantX Team.
- [IP-Adapter](https://github.com/tencent-ailab/IP-Adapter) and [ControlNet](https://github.com/lllyasviel/ControlNet). |