DragGAN, The Jouster's Review

The Essentials

• Image editing by manipulating control points (drag)
• Free, open source code available on GitHub
• From the Max Planck Institute of computer vision research
• Suited to AI researchers and developers exploring new generative editing interfaces

What is DragGAN?

DragGAN is a research prototype published by the Max Planck Institute that proposes a new way to edit generative images. The principle: you place points on a GAN-generated image and drag them to a target position. The model reconstructs the image following the movements, allowing you to change a character's pose, direction of gaze, or the shape of an object in a geometrically coherent way. The interface resembles a Bezier curve editor applied to image generation.

Strengths

Unprecedented interaction paradigm

The idea of "semantic drag and drop" is genuinely new. Rather than reformulating a prompt, you directly manipulate the image's geometry. For researchers exploring human-machine interfaces with generative models, it's a fertile line of work.

Open source and reproducible

The code is available on GitHub. Researchers can reproduce results, adapt the implementation and build on it. The method is fully transparent.

Convincing demo results

The demonstrations in the paper show impressive results on faces, cars and animals. The geometric coherence of transformations is difficult to achieve with other methods.

Limitations

This is not a product

DragGAN is an academic prototype, not a SaaS. There's no public web interface, no customer support, no regular updates. Installation requires Python, CUDA and a capable GPU.

Limited to GAN-generated images

DragGAN works with StyleGAN-generated images. It can't edit arbitrary photos or images produced by diffusion models like Stable Diffusion or Midjourney.

The ecosystem has evolved since publication

Since 2023, diffusion models have caught up and often surpassed GANs. Tools like Adobe Firefly or FLUX propose more versatile edits on arbitrary images.

Pricing

Free. Code is open source on GitHub (vcai.mpi-inf.mpg.de/projects/DragGAN). Local installation required, with associated GPU dependencies.

Alternatives

For more versatile image editing: Adobe Firefly. For pose manipulation: ControlNet. For diffusion editing: InvokeAI.

Verdict

DragGAN is an important academic contribution that opened a new path in interactive image editing. As a production tool in 2025, it's outdated. As a research object or inspiration for future interfaces, it remains a reference.

FAQ

Does DragGAN work on personal photos?

No, DragGAN requires images generated by the specific GAN it was trained on. It can't edit arbitrary photos.

Is there a web version of DragGAN?

Community-created Hugging Face Spaces demos exist but their availability is variable. There is no maintained official web version.

What GPU is needed to run DragGAN?

An NVIDIA GPU with at minimum 8 GB VRAM is recommended. An RTX 3080 or better offers reasonable generation times.

Is DragGAN integrated into commercial tools?

DragGAN's concepts have influenced features in tools like Adobe Firefly and Canva. DragGAN itself has not been commercialized directly.

Joute may earn a commission on subscriptions taken out via links in this article. This doesn't change our reviews.