AV ComfyUI Manual
32 / 64
Files
32
PART III · HANSEN BY TIMESTAMPS · 04:23

PEOPLE / PPL — a standalone production module, not “magic inside the master”

At this timestamp, Hansen moves to people. In this manual, the section acts as a bridge to two dedicated PEOPLE workflows: Generate & Place New People and Replace Existing People. The key here is to understand the shared module contract; implementation details are covered in the specialized chapters.

CONFIRMEDVideo review and source topology confirm separate PPL generation, segmentation, preparation, compositing and return stages. The detailed PEOPLE workflows are documented in dedicated chapters.
COMPOSITING

Two routes converge at selector 459

MODE 1FLUX person
477 Color → 449 Cut → 429 Paste → 672
459PPL Switchcurrent: image1
MODE 2Input / 3D inpaint
503/504 Cut → 494–496 Inpaint → 509 → 685
459Selected PPL
573Detail transferblend 0.09
67VAE Encode
57Main FLUX
53Decode
730LQ Saveactive output
HANSEN ORIGINAL

At 04:23, PEOPLE appears as an independent branch

The PPL branch does not begin by regenerating the entire architectural frame. Person generation and mask preparation exist as separate local stages, after which the result returns to the master scene.

The source route confirms a dedicated PPL prompt 408, person decode 829, Florence2/SAM2 mask chain, preparation, first composite 429 and return through selector 459.

ARCHVIZ FOUNDATION / EXTENSION

Read PEOPLE as one module contract

TWO SPATIAL CONTRACTS

The tools overlap, but the tasks are different

A common beginner mistake is to merge these two scenarios mentally. In the first, placement must be created. In the second, placement already exists and must be preserved.

WorkflowPlacement truthWhat AI does
01 · Generate & PlaceTarget region / placement inputCreates a new person and inserts them into the scene
02 · Replace ExistingExisting 3D/rendered personImproves quality/appearance while preserving the spatial slot
BEGINNER FOUNDATION

The key principle: a local AI module is safer than full-frame regeneration

When the task is local, a production workflow should give AI the smallest useful area of freedom. People are a good example: approved architecture remains fixed while generation focuses on the character and its integration.

PRACTICE

Exercise: find the boundaries of PEOPLE without reading its internal nodes

  • Find where PEOPLE receives its person source / prompt.
  • Find the first image checkpoint of the isolated person.
  • Find the segmentation stage.
  • Find the first composite with the architectural scene.
  • Find the selector/return after which PEOPLE becomes part of the master route again.