Abstract NeuroArt
Generate stylized morph videos and images StyleGAN3 is a model for generating the content, similar to what it has been trained on. It is used for replicating the specific style or object, and allows to manipulate images' features with vectors
14 styles
Perfect for abstract images and face generations
No input needed
512x512 output
High trainability
How to work with it
1. Choose dataset
Each dataset has unique data and covers a specific style or type of object. Choose it based on the result you want to get
2. Type any number as a seed
Seed is a result image of generation. With typing a number you choose what seed will be shown to you in the output image. This seed also will be used as the first and the last frame in the output video.
3. Choose number of keyframes
Number of keyframes correspond to how many seeds will be shown to you in the video output. Think of it like a frame interpolation between a set number of seeds
4. Launch
Press the green button to launch the AI magic process. You can download the results immediately, use 'Extract video frame' for getting a specific image out of the video and upscale it to achieve higher resolution
Input
Number
Output
Image, Video
Main task
Create images, create videos
Neural network
StyleGAN3
Release date
28.10.2021
License
Creative Commons CC BY-NC 4.0
Parameters
Dataset
A choice from 14 different models Each dataset has unique content and will give different results. We recommend to use • for abstract artworks (good for textures): complexity graphics, psychodelic, bundenko, nano world, landscape, gold and wikiart datasets • for avatars: Clonex dataset • for portraits: Face dataset • for cosmic objects: Hubble dataset • for cyborg character concept: Robots dataset
Seed
The range: from 1 to 9 000 Seed is a result image of generation. With typing a number you choose what seed will be shown to you in the output image. This seed also will be used as the first and the last frame in the output video. Each number = each unique image
Number of keyframes
The range: from 1 to 50 Number of keyframes corresponds to how many seeds will be in the video output. Number of keyframes = number of seeds in the generated video. The first image (seed) in the video sequence is taken from Seed. The number of keyframes defines how many seeds after this seed will be in the morph video We recommend to use 1 for quick generation of images Set 2 or higher to generate a morph video between several seeds
Duration (sec)
The range: from 1 to 50 It defines how long the video will be in seconds We recommend to set the number equal to the num_keyframes, so the video is not too slow or fast
FPS
Dropdown options: x15, x30, x60 and x120 FPS or frames-per-second defines how smooth the video will be. We recommend to use the standard 30 fps option
Images per column and row
The range: from 1 to 3 Batch generation, where all generated images and videos are combined into one file. It defines how many seeds will be shown in the final image and video output. Images per column corresponds with height or vertical seed placement, Images per row with width and horizontal placement Changing grid parameters will increase generation time and file size
Chaos
The range: from 0 to 5 It defines how far away from the dataset images the final result will be. The closer it is to 0, the more similarities to the dataset you will find. A higher number usually gives unexpected results, but also many artefacts We recommend to keep it between 0,2 to 1, so the result doesn't get too creepy
Work with attributes
A checkbox Turning it on will allow you to change how the seed looks by manipulating specific features and will produce the video output as a result In the process of training the neural network can determine specific features of the images (e.g., age) and assign a particular vector (represented as a number) to this feature. With this checkbox on you basically tell the neural network to show how this exact seed can look in different variations We recommend to use it only for 1 seed, meaning you have to set the parameter Number of keyframes to 1
Number of Attribute
The range: from 0 to 25 Choose which vector you want to use to change the image by typing a number. Each number from 0 to 25 is assigned to a particular vector. Each vector has a combination of features, according to which the image will be changed. Attributes are unique for each dataset For example: Number of attribute 1 for Faces dataset changes the gender from female to male, thins out the hair and sometimes adds glasses 5 Attribute changes the angle of the camera
Attribute strength
The range: from 0 to 50 It defines how noticeable the change of the vector and its features will be on the video The higher number is, the stronger the change will be, but it will also give more artefacts. We recommend to keep it between 3 and 6