The actor re-runs generation on an unchanged intent — the whole output, a selected part, or several variants at once — trading another draw for the work of fixing what came back.
Forms
Recurring forms: whole-output regenerate, partial regeneration of a selection (inpainting is the visual case), variant sets generated in parallel for comparison, and branches that keep the replaced result reachable.
Resources & references
- Shape of AI — Regenerate, Variations, Branches, Inpainting
- aiuxpatterns.com — Full Result Regeneration, Partial Regeneration, Result Variations
- AI UX Playground — Regeneration Carousel, Variation Picker, Output Comparison View
To-do
Generated content hosts the section-regeneration and variations stubs; migrate as this matures. Whether parallel variants deserve their own node is a fission question to revisit once examples accumulate.
Consequences
- the same intent yields another draw, so iteration happens without rearticulating the request — judgement cycles replace composition cycles
- drawing again discards silently: without branches or history the replaced result is simply gone, and the actor gambles what they have on getting something better
- parallel variants turn judgement into comparison, which is easier than absolute assessment and quietly caps how many drafts anyone will weigh
Related patterns
Enacts
- Agency — the actor disposes of output by re-asking rather than accepting what arrived
Complements
- Draft and publish — cheap draft modes make re-rolls affordable before committing to the full run
- Undo — restores what a regeneration replaced
Alternatives
- Instructed revision — naming the change versus drawing again on the same intent
- Editing in place — re-rolling versus hand-correcting the same output
Related
- Coordinated views — parallel variants are judged side by side
Preceded by
- Generated content — acts on the output the generation activity produced