인터페이스는 현지화되어 있으며, Skill 소개와 설명 문서는 선택한 언어로 AI가 자동 번역합니다. 코드·명령어·경로·전문 용어는 원문을 유지하며 원작자 문서가 우선합니다.
Skills 둘러보기
생산성 및 협업 · SKILL.md

draggan-ui

draggan-ui

Guides DragGAN desktop and Gradio point-based image manipulation, checkpoint selection, masks, latent-space controls, and UI troubleshooting.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill 문서

GitHub

name: draggan-ui description: "Guides DragGAN desktop and Gradio point-based image manipulation, checkpoint selection, masks, latent-space controls, and UI troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

DragGAN UI

Use this route for desktop visualizer or Gradio tasks: launch the app, load a StyleGAN checkpoint, choose a seed, place source/target point pairs, constrain edits with masks, or debug UI/model-loading failures.

Before launch

  1. Run the root scripts/check_environment.py with the local checkout and confirm CUDA for real drag optimization.
  2. Run the root scripts/check_model_assets.py --checkpoint-dir <dir>.
  3. Use references/ui-workflows.md for the control model and launch decisions.
  4. Use references/troubleshooting.md for GUI, checkpoint, CUDA, and import errors.

The bundled launch helpers are dry-run by default:

python sub-skills/draggan-ui/scripts/launch_draggan_gui.py \
  --repo-root /path/to/DragGAN --checkpoint-dir checkpoints
python sub-skills/draggan-ui/scripts/launch_gradio_demo.py \
  --repo-root /path/to/DragGAN --cache-dir checkpoints --listen

Pass --execute only after the printed command and preflight output are correct. The desktop helper scans existing .pkl files and supports --pkl, --capture-dir, and --browse-dir. The Gradio helper requires at least one checkpoint and supports --cache-dir, --listen, and --share.

Editing model

  • A point pair consists of a source point and a target point; incomplete pairs should be reset before starting.
  • The flexible-area mask allows optimization in selected regions; the fixed-area mode protects selected regions. Reset or show the mask before diagnosing a stalled drag.
  • w optimization is faster/more global; w+ is slower but usually gives finer local control.
  • Step size/learning rate, motion lambda, feature index, r1, and r2 affect stability and locality. Change one parameter at a time.
  • Loading a new checkpoint or seed resets image, point, and mask state.

Related routes

Bundled files

저장한 Skills