![[2] cd first-order-model / content/first-order-model 3) Click this button 5) Click the link Mount your Google drive folder on Colab [3] from google.colab import drive drive.mount ('/content/gdrive') 4) Click this button Go to this URL in a browser: https://accounts.google.com/o/oauth2/auth?client_id=9473189{ Enter your authorization code: .......... Mounted at /content/gdrive Add folder https://drive.google.com/drive/folders/1kZ1gCnpfU0BnpdU47pLM _TQ6RypDDqgw? usp=sharing to your google drive.](https://i.kym-cdn.com/photos/images/list/001/880/148/98d.jpg "Facemorphs / Baka Mitai Guide Image #2")
![+ Code + Text Copy to Drive E Files [22] from demo import load_checkpoints <> generator, kp_detector = load_checkpoints(config_path= checkpoint_path='/content/g first-order-model gdrive Perform image animation sample_data generated.mp4 make_animation ort img_as_ubyte Download Delete file ike_animation(source_image, driving_vide Rename file video Сoрy path 13) Right-click "generated.mp4, '../generated. mp4', [img_as_ubyte (frame Refresh wnloaded from /content folder then "download" HTML (display(source_image, driving_video, predictions).](https://i.kym-cdn.com/photos/images/list/001/880/043/e4a.jpg "Facemorphs / Baka Mitai Guide Image #10")
![Ofirst-order-model-demo.ipynb File Edit View Insert Runtime Tools Help Cannot save changes + Code + Text A Copy to Drive Table of contents Demo for paper "First Order Motion Model for Image Animation" [22] from demo import load_checkpoints generator, kp_detector = load_checkpo checkpoin <> Running on your data B Section Perform image animation 12) Click this folder icon from demo import make_animation from skimage import img_as_ubyte predictions = make_animation(source_i #save resulting video imageio.mimsave('../generated.mp4', [ #video can be downloaded from /conten HTML (display(source_image, driving_vi 100%|| 836/836 [00:34<00:00](https://i.kym-cdn.com/photos/images/list/001/880/042/04e.jpg "Facemorphs / Baka Mitai Guide Image #9")
![Create a model and load checkpoints [5] from demo import load_checkpoints generator, kp_detector = load_checkpoints (config_path='config/vox-256.yaml', 10) Press this button checkpoint_path='/content/gdrive/My Drive/first-order-motion-model/vox-cpk. Perform image animation [6] from demo import make_animation from skimage import img_as_ubyte 11) Then this one predictions = make_animation(source_image, driving_video, generator, kp_detector, relative=True) #save resulting video imageio.mimsave('../generated.mp4', [img_as_ubyte(frame) for frame in predictions]) #video can be downloaded from /content folder HTML (display(source_image, driving_video, predictions).to_htm15_video()) 100%| || 211/211 [00:07<00:00, 29.03it/s]](https://i.kym-cdn.com/photos/images/list/001/880/041/b25.jpg "Facemorphs / Baka Mitai Guide Image #8")
Facemorphs
Part of a series on Deepfakes. [View Related Entries]
[View Related Sub-entries]
This submission is currently being researched & evaluated!
You can help confirm this entry by contributing facts, media, and other evidence of notability and mutation.
About
Facemorphs, also known as First Order Motion Model Deepfakes, refer to videos using a motion capture method that makes static images of people's faces move according to the training model's motion. The method is similar to deepfakes, but uses only one still image instead of a set of images. Facemorphs gained popularity in July 2020 as they were applied to a variety of subjects, particularly to video game developer YandereDev. The method was used frequently with the Yakuza 0 song Baka Mitai.
In March 2021, facemorphs saw a surge in popularity following the release of WOMBO mobile app.
Origin
On February 29th, 2020, researchers from the University of Trento in Italy published a paper titled "First Order Motion Model for Image Animation." (FOMMIA, or FOMM) In the paper, the researchers describe a method for making a source image move according to the motion of a training model.[1]
On March 1st, content creator Verbalase uploaded an Instagram[2] video on Instagram using images of people such as Barack Obama and Donald Trump (shown below). It is unknown whether Verbalase used FOMMIA although but it is possible given how close the upload date is to the paper's publication. The YouTube reupload of Verbalase's video posted on March 8th received over 530,000 views in four months.[3]
Facemorphing was popularized by content creator @heyitswindy, who on July 9th, 2020, tweeted[4] one of the earliest examples of a humorous facemorph based on a popular photograph of YandereDev and the sound from Consume the Cum Chalise video.
deepyandere fakedev pic.twitter.com/tLYIEo9t6b
— Windy (@heyitswindy) July 9, 2020
Spread
On July 10th, 2020, @heyitswindy[5] posted a similar facemorph of YandereDev using audio from a recent TikTok, receiving almost 50,000 likes and 11,700 tweets (shown below). In a reply to the tweet, @heyutwindy mentioned that he used a demo for FOMMIA to create the images.[6]
— Windy (@heyitswindy) July 10, 2020
In a July 13th tweet, @heyitswendy demonstrated the use of the software method with an image of YandereDev (shown below).
The man behind the slaughter. pic.twitter.com/Zre6I5eBdX
— Windy (@heyitswindy) July 13, 2020
Starting in mid-July 2020, Facemorphs gained significant popularity on Twitter, with memes ofter using photographs of internet-popular people and characters.
On July 19th, heyitswindy uploaded a tutorial video for doing facemorphs to his YouTube channel, gaining over 80,000 views in less than two weeks (shown below).[7]
WOMBO
Wombo, stylized as WOMBO and also known as Wombo.ai, is a mobile application that generates lip sync facemorphs based on a provided image and using a selection of available songs, including "Chug Jug With You", "Baka Mitai," and "What Is Love?". Launched in January 2021, the app achieved viral popularity in March 2021.
Leaked footage from the @Tesla Q3 Earnings Call 🤫#TopSecret pic.twitter.com/QUHcrhXKM2
— WOMBO (@WOMBO) October 26, 2020
Various Examples
— Cakewalking (@Tortokhod) July 21, 2020
— SiveOrSimp 1️⃣ ➐ (@sivemorten) July 27, 2020
External References
[1] Aliaksandr Siarohin, Stéphane Lathuilière, Sergey Tulyakov, Elisa Ricci, Nicu Sebe – First Order Motion Model for Image Animation
[3] YouTube Obama Beatbox by Verbalase
[4] Twitter – @heyitswindy
[5] Twitter – @heyitswindy
[6] Aliaksandr Siarohin – "Demo for paper First Order Motion Model for Image Animation
f7. YouTube – heyitswindy
![Load driving video and source image * V GD 0 i 7 import imageio import numpy as np import matplotlib.pyplot as plt import matplotlib.animation as animation 9) Press this button from skimage.transform import resize from IPython.display import HTML import warnings warnings.filterwarnings("ignore") source_image = imageio.imread('/content/gdrive/My Drive/first-order-motion-model/02.png') driving_video = imageio.mimread('/content/gdrive/My Drive/first-order-motion-model/04.mp4') #Resize image and video to 256x256 source_image = resize(source_image, (256, 256))[..., :3] driving_video = [resize(frame, (256, 256))[..., :3] for frame in driving_video] def display(source, driving, generated=None): fig = plt.figure (figsize=(8 + 4 * (generated is not None), 6)) ims = [] for i in range(len(driving)): cols = [source] cols.append (driving[i]) if generated is not None: cols.append(generated[i]) im = plt.imshow(np.concatenate(cols, axis=1), animated=True) plt.axis ('off') ims.append([im])](https://i.kym-cdn.com/photos/images/list/001/880/040/41f.jpg "Facemorphs / Baka Mitai Guide Image #7")
![Demo for paper "First Order Motion Model for Image Animation" 1) Paste this line Clone repository !pip install PYYAML==5.3.1 !git clone https://github.com/AliaksandrSiarohin/first-order-model Cloning into 'first-order-model'... remote: Enumerating objects: 85, done. remote: Total 85 (delta 0), reused e (delta 0), pack-reused 85 Unpacking objects: 100% (85/85), done. 2) Press this button [2] cd first-order-model A / content/first-order-model Mount your Google drive folder on Colab](https://i.kym-cdn.com/photos/images/list/001/880/032/a43.jpg "Facemorphs / Baka Mitai Guide Image #1")
Top Comment
MrBoom420
Jul 30, 2020 at 01:10PM EDT