Workflow Design

Preserve The Baseline First

Workflow iteration gets risky when the original structure is lost. Before large changes:

• save an untouched upstream workflow backup
• keep milestone snapshots for each structural revision
• compare against the upstream graph before introducing helper nodes
• treat compatibility shims and native rewrites as separate choices

Choose The Input Model Early

The public workflow surface should match how the user thinks:

• use a folder selector when the workflow should consume a variable number of images
• use an upload widget when the user naturally replaces one file at a time
• avoid hard-coding multiple upload slots when the real need is "pick from a set"
• choose a model the user can explain in one sentence

Prefer A Native Public Node Surface

Once the canonical node names, categories, and widget labels are clear:

• standardize sample workflows on that surface
• keep old compatibility layers explicit and isolated
• avoid hiding major workflow drift behind convenience wrappers

Keep Sample Workflows Shareable

• do not ship machine-local paths in a workflow meant for others
• keep defaults blank for public examples
• inject sample assets only during automated smoke runs
• separate heavy production graphs from lightweight smoke workflows

Treat Layout As Maintainability

Readable graphs are easier to debug and review. Watch for:

• overlapping groups or nodes
• title-band collisions
• nodes falling outside intended group bounds
• App mode metadata drifting away from actual widget layout

Release Checklist

• explain the App mode input model in the published docs
• keep install or clone steps short enough for ephemeral environments
• publish sample workflows that are safe to share and safe to validate automatically