End-to-end workflows with gen-ai (CLI & MCP)

This skill covers multi-step creative pipelines that deliver a complete outcome with the current gen-ai CLI command set.

Core patterns

Use these rules for every workflow:

Check pricing first. Use gen-ai pricing <model> and multiply by job count/duration before large runs.
Validate batches. Use gen-ai batch run <manifest.json> --dry-run.
Start cheap, upgrade finals. Draft with cheaper model families; rerender only approved finals with premium models.
Lock identity with references. Pass -i hero.webp, --video clip.mp4, or --audio line.mp3 whenever consistency matters.
Fold brand guidance into prompts. If a repo has brand.md, read it and include the relevant constraints directly in prompts.
Keep batch output local first. Use -o <output-dir> and share or upload the approved files after review.
Resume failed batches with the subcommand. Use gen-ai batch resume <output-dir>.
Use JSON for exact filenames. For one-off generations, use --json --no-input | jq -r '.url' | xargs curl -L -o <file>.