My cel-shade ComfyUI workflow - flat color pass + separate stabilized line pass (json + sample)

by @jordansingh in c/anime-lab · · Workflow

My cel-shade ComfyUI workflow - flat color pass + separate stabilized line pass (json + sample)
The graph from my cel-shading post, cleaned up and shared. Two parallel branches: a flat-color branch (with the shadow-quantize node) and a lineart branch (ControlNet lineart, rendered separately), composited at the end. MIT.
github.com

The graph from my cel-shading post, cleaned up and shared. Two parallel branches: a flat-color branch (with the shadow-quantize node) and a lineart branch (ControlNet lineart, rendered separately), composited at the end. MIT.

501 points · 7 comments

Comments

  • @skylerlima_creates · · 61 points

    The parallel flat-branch + line-branch architecture is the correct mental model for anime and seeing it as an actual node graph makes it click. Color and line are SEPARATE departments that meet at compositing, just like a real studio's shiage and douga.

  • @chenwilliams_studio · · 48 points

    OP - I literally named the groups 'douga' and 'shiage' in the graph as a joke but it is also just accurate, we are rebuilding the anime production pipeline one node group at a time.

  • @chenthompson_fx · · 29 points

    Does the shadow-quantize node choke on complex backgrounds or is it character-only? My flats look great on the character and the background goes to posterized mud.

  • @oscarjohnson_films5727 · · 34 points

    OP - character-only, yeah. Backgrounds want a totally different treatment (they are usually PAINTED in anime, not cel-shaded). Mask the character, quantize only them, treat the BG as a separate matte-painting pass. Different departments again.

  • @dmitrirodriguez_ai · · 24 points

    MIT + a working json + it does what the post claimed, exemplary. starred and stealing the quantize node for my own graph.