---

name: stylegan-generation description: "Guides DragGAN and StyleGAN-Human batch generation, interpolation, style mixing, video command planning, and legacy pickle conversion." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

StyleGAN Generation

Use this route for non-interactive generation from StyleGAN2/3 pickles, StyleGAN-Human generation, latent interpolation, style-mixing grids/videos, seed/truncation planning, or legacy pickle conversion.

Command planning

Use the bundled command builder; it prints commands and never runs model inference:

python sub-skills/stylegan-generation/scripts/build_generation_command.py \
  --repo-root /path/to/DragGAN gen-images \
  --network checkpoints/model.pkl --seeds 1,2,5-10 --outdir outputs/generated

Read references/generation-workflows.md for each workflow and references/model-loading-and-conversion.md for checkpoint formats and naming. Use references/troubleshooting.md before retrying a failed command.

Main routes

• Top-level generation: seed ranges, truncation, noise mode, conditional class labels, translation, rotation, and output image names.
• StyleGAN-Human generation: choose version 1/2/3 explicitly; v2/v3 use the PyTorch path, while v1 needs a TensorFlow 1.x-compatible environment.
• Interpolation: exactly two useful seeds, GIF frame count, FPS, and optional intermediate frames.
• Style mixing: row/column seed grids and style-layer ranges.
• Style-mixing video: expensive and import-gated by the source’s TensorFlow dnnlib.tflib import; use only after checking that dependency.
• Legacy conversion: convert supported TensorFlow-era StyleGAN2/StyleGAN2-ADA pickles into the native PyTorch pickle format.

Related routes

• Interactive point editing: ../draggan-ui/SKILL.md.
• StyleGAN-Human alignment, PTI, attribute editing, and InsetGAN: ../stylegan-human-manipulation/SKILL.md.
• Training: ../stylegan-training/SKILL.md.

Bundled files

• scripts/build_generation_command.py builds exact option names without executing.
• references/generation-workflows.md contains task recipes and output expectations.
• references/model-loading-and-conversion.md contains model and conversion rules.
• references/troubleshooting.md contains generation-specific recovery.