Snap Inc.

Pavel Savchenkov

Aleksandr Mashrabov

Arnab Ghosh

Sergey Smetanin

Sergey Tulyakov

Jian Ren

Alexey Pchelnikov

Dmitry Matov

Aleksandr Zakharov

Eugene Krokhalev

Aleksandr Belskikh

Jeremy Baker Voss

Daniel Moreno

Menglei Chai

Roman Golobokov

Nikita Demidov

Ivan Babanin

Maxim Lukin

Yurii Volkov

Nikita Belosludtcev

Anna Kovalenko

Vladimir Gordienko

Dor Ayalon

Timur Zakirov

Sofya Savinova

Nikolai Smirnov

Antoine Chassang

Alexey Pankov

Georgii Grigorev

Mykyta Bakunov

Hee Hun Kim

Aliaksei Mikhailiuk

Neha Yadav

Bingqian Lu

Tianxiang Gao

Examples disclosed herein describe techniques for automatic image quality evaluation. A first set of images generated by a first automated image generator and a second set of images generated by a second automated image generator are accessed. A first machine learning model generates a first quality indicator for each image in the first set of images and the second set of images. A second machine learning model generates a second quality indicator for each image in the first set of images and the second set of images. Based on the generated indicators, a first image from the first set of images and a second image from the second set of images are automatically selected and compared. A first ranking of the first automated image generator and the second automated image generator is generated based on the comparison, and ranking data is caused to be presented on a device.

Automatic image quality evaluation

Described is a system for generating a textual response from a received image by determining participation in an interaction function by a first user of an interaction system, identifying an image associated with the participation, processing data associated with the image using a first machine learning model to identify one or more features within the image, and generating a prompt based on the identified one or more features. The system then identifying instructions for a second machine learning model, processing the prompt and the instructions using the second machine learning model to generate a textual response to the image, and causing display of the textual response within the interaction function to the first user.

Image-to-text large language models (LLM)

A method of generating an image for use in a conversation taking place in a messaging application is disclosed. Conversation input text is received from a user of a portable device that includes a display. Model input text is generated from the conversation input text, which is processed with a text-to-image model to generate an image based on the model input text. The coordinates of a face in the image are determined, and the face of the user or another person is added to the image at the location. The final image is displayed on the portable device, and user input is received to transmit the image to a remote recipient.

Text-guided cameo generation

The subject technology receives an input image and a segmentation mask of the input image. The subject technology obtains reconstructed noise of the input image using the input image and the segmentation mask. The subject technology determines a first set of features by performing a first portion of a forward pass of the reconstructed noise through a decoder. The subject technology determines a second set of features by processing the input image for stable diffusion using an image to image (IMG2IMG) model. The subject technology generates a third set of features based on combining, using the segmentation mask, the first set of features and the second set of features with the reconstructed noise. The subject technology generates an output image by performing a remaining portion of the forward pass of the third set of features through the decoder.

Face identity preservation for image-to-image models using stable diffusion generative model

The subject technology receives frames of a source media content, the frames of the source media content including representations of a head and a face of a source actor. The subject technology generates, based at least in part on the frames of the source media content, sets of source pose parameters. The subject technology receives at least one target image, the at least one target image including representations of a target head and a target face of a target entity. The subject technology provides the sets of source pose parameters to a neural network to determine facial landmarks for head turns and facial expressions. The subject technology generates, based at least in part on the sets of source pose parameters and the facial landmarks for head turns and facial expressions, an output media content. The subject technology provides augmented reality content based at least in part on the output media content for display on a computing device.

Face detection to generate parameters for a neural network that determines facial landmarks used for facial synthesis for head turns in augmented reality content

Provided are systems and methods for face reenactment. An example method includes receiving a target video that includes at least one target frame, where the at least one target frame includes a target face, receiving a scenario including a series of source facial expressions, determining, based on the target face, a target facial expression of the target face, synthesizing, based on a parametric face model and a texture model, an output face including the target face, where the target facial expression of the target face is modified to imitate a source facial expression of the series of source facial expressions, and generating, based on the output face, a frame of an output video. The parametric face model includes a template mesh pre-generated based on historical images of faces of a plurality of individuals, where the template mesh includes a pre-determined number of vertices.

Face reenactment

Systems and methods for text and audio-based real-time face reenactment are provided. An example method includes receiving an input text and a target image, where the target image including a target face, generating, based on the input text, a sequence of sets of acoustic features corresponding to the input text, generating, based on the sequence of sets of acoustic features, a sequence of sets of mouth key points, generating, based on the sequence of sets of mouth key points, a sequence of sets of facial key points, determining, based on the set sequence of sets of facial key points, a sequence of deformations of the target face, and applying the sequence of deformations to the target image, thereby generating a sequence of frames of an output video.

Text and audio-based real-time face reenactment

Embodiments enable virtual hair generation. The virtual hair generation can be performed by generating a first image of a face using a GAN model, applying 3D virtual hair on the first image to generate a second image with 3D virtual hair, projecting the second image with 3D virtual hair into a GAN latent space to generate a third image with virtual hair, performing a blend of the virtual hair with the first image of the face to generate a new image with new virtual hair that corresponds to the 3D virtual hair, training a neural network that receives the second image with the 3D virtual hair and provides an output image with virtual hair, and generating using the trained neural network, a particular output image with hair based on a particular input image with 3D virtual hair.

Generating virtual hairstyle using latent space projectors

Provided are systems and methods for portrait animation. An example method includes receiving, by a computing device, scenario data including information concerning movements of a first head, receiving, by the computing device, a target image including a second head and a background, determining, by the computing device and based on the target image and the information concerning the movements of the first head, two-dimensional (2D) deformations of the second head in the target image, applying, by the computing device, the 2D deformations to the target image to obtain at least one output frame of an output video, the at least one output frame including the second head displaced according to the movements of the first head, and filling, by the computing device and using a background prediction neural network, a portion of the background in gaps between the displaced second head and the background.

Photorealistic real-time portrait animation

Examples disclosed herein describe techniques related to automated image generation in an interaction system. An image generation request is received from a first user device associated with a first user of an interaction system. The image generation request comprises a text prompt. Responsive to receiving the image generation request, an image is automatically generated by an automated text-to-image generator, based on the text prompt. The image is caused to be presented on the first user device. An indication of user input to select the image is received from the user device. Responsive to receiving the indication of the user input to select the image, the image is associated with the first user within the interaction system, and a second user of the interaction system is enabled to be presented with the image.

Automatic image generation in an interaction system

Examples disclosed herein describe prompt modification techniques for automated image generation. An image generation request comprising a base prompt is received from a user device. A plurality of prompt modifiers is identified. A processor-implemented scoring engine determines, for each prompt modifier, a modifier score. The modifier score for each prompt modifier is associated with the base prompt. One or more of the prompt modifiers are automatically selected based on the modifier scores. A modified prompt is generated. The modified prompt is based on the base prompt and the one or more selected prompt modifiers. The modified prompt is provided as input to an automated image generator to generate an image, and the image is caused to be presented on the user device.

Prompt modification for automated image generation

A methodology for training a machine learning model to generate color-neutral input face images is described. For each training face image from a training dataset that is used for training the model, the training system generates an input face image, which has the color and lighting of a randomly selected image from the set of color source images, and which has facial features and expression of a face object from the training face image. Because, during training, the machine learning model is 'confused' by changing the color and lighting of a training face image to a randomly selected different color and lighting, the trained machine learning model generates a color neutral embedding representing facial features from the training face image.

Face animation synthesis

A method of generating an image for use in a conversation taking place in a messaging application is disclosed. Conversation input text is received from a user of a portable device that includes a display. Model input text is generated from the conversation input text, which is processed with a text-to-image model to generate an image based on the model input text. The generated image is displayed on the portable device, and user input is received to transmit the image to a remote recipient.

Text-guided sticker generation

Provided are systems and methods for portrait animation. An example method includes receiving, by a computing device, a scenario video, where the scenario video includes information concerning a first face, receiving, by the computing device, a target image, where the target image includes a second face, determining, by the computing device and based on the target image and the information concerning the first face, two-dimensional (2D) deformations of the second face in the target image, and applying, by the computing device, the 2D deformations to the target image to obtain at least one output frame of an output video.

Provided are systems and methods for realistic head turns and face animation synthesis. An example method includes receiving a source frame of a source video, where the source frame includes a head and a face of a source actor, generating source pose parameters corresponding to a pose of the head and a facial expression of the source actor; receiving a target image including a target head and a target face of a target person, determining target identity information associated with the target head and the target face of the target person, replacing source identity information in the source pose parameters with the target identity information to obtain further source pose parameters, and generating an output frame of an output video that includes a modified image of the target face and the target head adopting the pose of the head and the facial expression of the source actor.

Growing community of inventors

London, United Kingdom

Pavel Savchenkov