---

name: python-api description: "Routes DragGAN scripting tasks such as loading checkpoints, generating latents and images, and running the drag optimization loop from Python." disable-model-invocation: true metadata: disco-role: operating license: NO_LICENSE

python-api

Use this sub-skill when the task is about draggan.draggan or draggan.utils: loading a model, generating a W latent, synthesizing an image, preparing points, or writing a script around the drag loop.

Start here

1. Run the bundled preflight: ../../scripts/check_install.py --mode api
2. Read references/api-reference.md for the verified function signatures and return shapes.
3. Read references/workflows.md for a minimal scripted generation or drag loop.
4. Read ../../references/checkpoints.md if you need checkpoint names or cache behavior.
5. Read references/troubleshooting.md if the tensor or device setup is failing.

This sub-skill covers

• Loading a pretrained generator with load_model().
• Creating W+ latents with generate_W().
• Turning a latent back into an image with generate_image().
• Running the iterative drag loop with drag_gan() and point tracking.
• Using the point and mask helpers in draggan.utils.

This sub-skill does not cover

• Browser UI layout or Gradio launch flags.
• Docker packaging or checkpoint catalog browsing.
• Claims that the drag loop works on CPU or MPS as a verified path.

Important cautions

• The verified drag loop is CUDA-only in this snapshot.
• Points are represented as [y, x] tensors on the same device as the model.
• The mask parameter is accepted by the current drag loop signature but is not enforced by the current implementation.
• Conditional checkpoints need an explicit class_idx; unconditional checkpoints ignore it.

Helpful references

• references/api-reference.md for the verified signatures and shape notes.
• references/workflows.md for copyable Python recipes.
• references/troubleshooting.md for device, point-order, and mask mistakes.
• ../../references/checkpoints.md for checkpoint names and the cache root.