L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Développement · SKILL.md

web-demo

web-demo

Routes DragGAN browser-demo tasks such as launching the Gradio UI, selecting checkpoints, dragging points, and saving edited outputs.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: web-demo description: "Routes DragGAN browser-demo tasks such as launching the Gradio UI, selecting checkpoints, dragging points, and saving edited outputs." disable-model-invocation: true metadata: disco-role: operating license: NO_LICENSE

web-demo

Use this sub-skill when the task is about the browser experience: launch the demo, open the web UI, choose a checkpoint, click handle and target points, save an image or video, or troubleshoot the demo runtime.

Start here

  1. Run the bundled preflight: ../../scripts/check_install.py --mode web
  2. Launch the demo with the bundled wrapper: ../../scripts/launch_web_demo.py --device cuda --ip 0.0.0.0 --port 7860
  3. Read references/workflows.md for the point-drag sequence and UI controls.
  4. Read ../../references/checkpoints.md if you need model names or cache behavior.
  5. Read references/troubleshooting.md if the UI opens but the interaction fails.

This sub-skill covers

  • Launching the browser demo with the supported launch flags.
  • Switching between checkpoints in the model dropdown.
  • Creating a seeded image, placing handle and target points, and running Drag it.
  • Resetting, undoing, and saving edited outputs.
  • Understanding the saved draggan_tmp/ files and checkpoint cache usage.

This sub-skill does not cover

  • Writing a custom Python script around draggan.draggan functions.
  • Low-level tensor shapes, point encoding, or latent manipulation details.
  • Claims that CPU or MPS is a verified drag backend.

Important cautions

  • The verified drag path in this snapshot requires CUDA.
  • The Draw a Mask tab is present in the UI, but the current optimizer does not enforce the mask.
  • Uploaded-image inversion is experimental; treat it as a best-effort path rather than the default workflow.
  • If you only need the install check or environment diagnosis, use the root scripts and troubleshooting reference first.

Helpful references

  • references/workflows.md for the user-facing browser steps.
  • references/troubleshooting.md for launch, model-download, and UI-specific failures.
  • ../../references/deployment.md for install and launch modes shared with the root skill.
  • ../../references/checkpoints.md for checkpoint names and the cache root.
Skills enregistrées