comfyui_gr85 – Custom Nodes for ComfyUI

This repository provides a small collection of GR85-flavoured custom nodes for ComfyUI. It includes utilities for working with image dimensions, randomization, prompt manipulation, and seed utilities.

The implementation is being migrated from the legacy INPUT_TYPES / RETURN_TYPES API to the modern comfy_api.latest schema-based API (io.ComfyNode, define_schema, execute, and ComfyExtension).

Currently Simple Wildcard Picker has been migrated fully to the new API; other nodes still use the classic style.

---

Node overview

Resolution helpers

• GR85_ImageDimensionResizer (ImageDimensionResizer, category GR85/Resolution)

  • Computes new width/height based on a selected target resolution while preserving the original pixel count and aspect ratio.

• GR85_ImageSizer (ImageSizer, category GR85/Resolution)

  • Rescales to a target aspect ratio with orientation (original / landscape / portrait) and snaps the result to a given tolerance (e.g. multiples of 16).

• GR85_ImageSizerAll (ImageSizerAll, category GR85/Resolution)

  • Takes a pixel budget and aspect ratio components and returns new dimensions that match the pixel count and orientation, adjusted to a given tolerance.

• GR85_RandomRatio (RandomRatio, category GR85/Resolution)

  • Given two width/height pairs, samples a random aspect ratio between their ratios and returns it as integer width/height components.

Prompt helpers

• GR85_SeedBasedOutputSelector (SeedBasedOutputSelector, category GR85/Prompt/Selection)

  • Selects one string from up to 10 inputs based on a seed. Only non-None inputs are considered, and the seed is used with modulo to choose the index.

• GR85_SimpleWildcardPicker (SimpleWildcardPicker, category GR85/Prompt/Wildcards)

  • New implementation (v3 API):
    • Implemented as an io.ComfyNode using define_schema and execute.
    • Inputs:
      • prompt (multiline string)
      • seed (integer)
    • Behavior: expands {a|b|c}-style wildcards (with support for nested braces) using a seeded RNG so results are reproducible.

• Tag injector nodes (all classic API for now, category GR85/Prompt/Tags)

  • TagInjectorSingle – injects a single tag into a template using placeholder names like __elements__.
  • TagInjectorDuo – same idea, but for two tags (e.g. elements and stuff).
  • TagInjectorLarge – larger template that injects multiple semantic placeholders such as __location__, __weather__, __style__, etc.

Random / utility nodes

• GR85_NextSeed (NextSeed, category GR85/Random/Seed)

  • Given a seed, produces a new random seed in the full 64‑bit range.

• GR85_RandomFloat (RandomFloat, category GR85/Random/Numbers)

  • Generates a random float in [min_value, max_value], with configurable decimal precision, seeded for reproducibility.

• GR85_RandomInt (RandomInt, category GR85/Random/Numbers)

  • Generates a random integer in [min_value, max_value], seeded for reproducibility.

---

Architecture and loading

This extension supports both:

• Legacy discovery via NODE_CLASS_MAPPINGS / NODE_DISPLAY_NAME_MAPPINGS in __init__.py (for most nodes).
• Modern discovery via comfy_api.latest and a ComfyExtension entrypoint (for SimpleWildcardPicker).

The root __init__.py provides:

• NODE_CLASS_MAPPINGS and NODE_DISPLAY_NAME_MAPPINGS for the classic nodes.
• GR85Extension(ComfyExtension) with:
  • get_node_list() returning the list of new io.ComfyNode nodes.
• async def comfy_entrypoint() which ComfyUI calls to load the extension.

---

Migration guide

1. Simple Wildcard Picker

What changed

• Previously, SimpleWildcardPicker was a classic custom node defined with:
  • INPUT_TYPES
  • RETURN_TYPES, RETURN_NAMES, FUNCTION, CATEGORY
  • Registration via NODE_CLASS_MAPPINGS["GR85_SimpleWildcardPicker"] in __init__.py.
• That legacy implementation and registration have been removed.
• A new implementation now exists:
  • Class: SimpleWildcardPicker(io.ComfyNode) in nodes/prompt/simple_wildcard_picker.py.
  • Schema: define_schema() using io.Schema, io.String.Input, io.Int.Input, io.String.Output.
  • Execution: execute() returns an io.NodeOutput.
  • Registration: via GR85Extension.get_node_list() and comfy_entrypoint().
  • Node ID: still GR85_SimpleWildcardPicker (no v3 suffix), display name "Simple Wildcard Picker".

Behavior compatibility

• The wildcard expansion logic has been preserved:
  • Same {a|b|c} syntax.
  • Nested braces are supported.
  • A seeded RNG (Random(seed)) is used for reproducible choices.
• The internals were refactored into a helper function (_process_wildcards) used by the new execute() method.

Impact on existing workflows

• Graphs that reference node ID GR85_SimpleWildcardPicker should continue to work, but they now use the new schema-based node.

• Code or tooling that relied on:

  • NODE_CLASS_MAPPINGS["GR85_SimpleWildcardPicker"], or
  • the legacy INPUT_TYPES/FUNCTION shape of SimpleWildcardPicker

  will need to be updated to work with the new io.ComfyNode API and extension-based discovery.

How to update custom integrations

• Instead of importing and introspecting the legacy class, treat it as a standard io.ComfyNode:
  • Use its schema via SimpleWildcardPicker.define_schema().
  • Expect execute(prompt, seed) to be a classmethod returning io.NodeOutput.
• Do not rely on NODE_CLASS_MAPPINGS for this node anymore; it is no longer present there.

2. Other nodes (classic API)

• All other GR85 nodes still use the classic:
  • INPUT_TYPES
  • RETURN_TYPES / RETURN_NAMES
  • FUNCTION / instance methods
  • NODE_CLASS_MAPPINGS / NODE_DISPLAY_NAME_MAPPINGS
• They will continue to behave as before.

Planned direction

• Over time, these nodes can be migrated in the same style as SimpleWildcardPicker:
  • Introduce an io.ComfyNode implementation with define_schema + execute.
  • Register it via GR85Extension.get_node_list().
  • Optionally remove the old mapping entry once you are comfortable with the migration.

If you want to migrate another specific node (e.g. GR85_SeedBasedOutputSelector or one of the latent utilities), follow the same pattern as SimpleWildcardPicker and update the extension’s get_node_list() accordingly.