OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Ganspace

GANSpace is an open-source research codebase for discovering interpretable controls in generative adversarial networks and using them to edit generated images. It applies principal-component analysis to model activations to identify latent directions, with layer-wise edits demonstrated across multiple GANs.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Ganspace — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use GANSpace when you want to investigate how a supported GAN’s latent directions correspond to visual changes, or to apply discovered directions as image edits.

Who it is for

  • Researchers studying GAN representations and image synthesis.
  • Developers and practitioners who want to explore or visualize latent directions in supported GAN models.

Documented uses

  • Explore controls in BigGAN, StyleGAN and StyleGAN2 models.
  • Apply previously saved edits interactively.
  • Visualize principal components, including producing videos for a documented StyleGAN example.
  • Recreate figures from the associated paper using included Jupyter notebooks.
  • Integrate a new model by implementing the repository’s BaseModel wrapper interface and registering it.
Before you choose

Requirements & limits

The public GitHub repository lists an Apache-2.0 license; the README identifies Python 3.7 and PyTorch 1.3 and links a Google Colab notebook. It notes that some included directories use MIT licenses.

  • The documented workflow centers on particular GAN models, classes, layers and latent spaces rather than a model-independent image editor.
  • The README describes a command-line interactive viewer and visualization scripts, as well as Jupyter notebooks for paper figures.
  • Adding a model requires a wrapper using the BaseModel interface and registering it in get_model().
  • The README reports an interactive-viewer startup freeze sometimes occurring on Ubuntu 18.04.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Explore a model’s GAN controls from the command line

Run interactive exploration for one of the repository’s documented GAN model and class combinations.

Before you start

  • The GANSpace repository and its setup instructions.
  • Python 3.7 and PyTorch 1.3 are identified in the README.
  • A supported model, class and layer; examples are provided for BigGAN-512 and StyleGAN2.

Steps

  1. Follow the repository’s SETUP.md instructions.

  2. Choose a documented model, class and layer combination.

  3. Run interactive.py with the corresponding options; for example, the README shows BigGAN-512 with class husky and layer generator.gen_z.

  4. Use the interactive exploration to examine controls discovered from the model.

Expected outcome

Interactive exploration of PCA-derived controls for the selected GAN configuration.

Guide limitations

  • The README warns that the interactive viewer sometimes freezes at startup on Ubuntu 18.04.
  • The repository’s setup instructions are required; this guide does not reproduce them.
Documentation 1 ↗
Source-based practical guide

Visualize StyleGAN principal components

Generate a visualization of principal components for a documented StyleGAN model configuration.

Before you start

  • The GANSpace repository and its setup instructions.
  • A StyleGAN or StyleGAN2 model configuration supported by the repository.

Steps

  1. Follow the repository’s SETUP.md instructions.

  2. Choose the model, class, latent space and layer to visualize.

  3. Run visualize.py with those options; the README shows StyleGAN2, class ffhq, W space and layer style as an example.

  4. For the documented StyleGAN wikiart example, add --batch --video to create videos saved to ./out.

Expected outcome

A principal-component visualization, or video output for the documented batch-video example.

Guide limitations

  • The README provides example configurations but does not establish that every model checkpoint is included or available.
  • The setup instructions are required for execution.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
harskish/ganspace · Official homepage title · retained in its source language
How we review sourcesSuggest a correction