Rebalance Pack: ComfyUI Nodes for Krea 2 and Ideogram 4 Image Editing

What Rebalance Pack is for and who it is aimed at

Rebalance Pack is a ComfyUI custom node collection, written in Python, distributed under Apache-2.0. The README describes it as "a set of evolving nodes that streamline the best use practices in ComfyUI", and the pyproject.toml carries a longer description that splits the nodes into four groups: foundational, utility, conditioning, and model specific.

The audience is narrow and identifiable. The model-specific group names exactly two families: Krea 2 and Ideogram 4, with nodes such as Conditioning Krea 2 Rebalance, Krea 2 Edit Rebalance, Krea 2 Encode Rebalance, and their Ideogram 4 counterparts. If you are not running those models in ComfyUI, most of the pack's distinguishing surface does not apply to you. The foundational and utility groups (Omni Node, Input, Load Images, Concatenate (Multi), String (Inline), Any, Switch, plus resolution capping, aspect-ratio cropping, uncropping and a border mask detector) are generic enough to be useful in any graph, but they are not what the release notes lead with.

The README is written as a changelog with dates, not as an introduction. Entries run from 29/06 through 12/07, 29/07 and 10/08, tracking features like multi-character references, negative prompting for "stripping undesired effects", and the Omni Node family. A new user has to reconstruct the current feature set from these entries, because there is no single page that lists what each node accepts and returns.

How the Omni Node rebuilds its own inputs and outputs

The mechanism the README describes most concretely is the Omni Node, introduced in the 29/07 entry: "A sole ComfyUI node that lets you paste in the body of a node class, rebuilding its own inputs/outputs to mirror the source."

That is a metaprogramming approach rather than a fixed node schema. Normally a ComfyUI node declares its INPUT_TYPES and RETURN_TYPES statically, and the graph editor draws sockets from that declaration. Omni Node instead takes the body of a node class as text and derives the socket layout from it at runtime. The practical consequence is that one node can stand in for many shapes of graph without you writing a custom node file or restarting ComfyUI with a new registration. The trade-off is equally clear: the graph you build is only as correct as the class body you paste, and nothing in the README describes validation, error messages, or what happens when the pasted body references a type the node cannot resolve.

The rest of the pack follows a conventional ComfyUI layout. The repository root holds one Python module per concern: omni_nodes.py, foundational.py, conditioning_rebalance.py, krea2.py, ideogram4.py, mask_toolkit.py, model_toolkit.py, resize_toolkit.py, utilities.py, with __init__.py as the entry point ComfyUI imports. There is a web/ directory for front-end assets and a workflows/ directory holding example graphs, including the Krea 2 Edit (Full Context) workflow linked from the 12/07 entry.

Installing Rebalance Pack into ComfyUI

The README gives one installation command, a clone of the repository. It does not document a package index install, a requirements file, or a pinned ComfyUI version. Run the clone from inside your ComfyUI custom_nodes directory so the folder lands where ComfyUI scans for custom nodes.

cd ComfyUI/custom_nodes
git clone https://github.com/nova452/Rebalance-Pack.git

After the clone, restart ComfyUI. The pyproject.toml registers the pack with ComfyUI under the publisher id nova452 and the display name "Rebalance Pack", with avatar.svg as its icon, so the nodes should appear under that name in the node menu once the import succeeds. The project metadata also points to a wiki at the repository's /wiki path and to the issues tracker for bug reports; the README itself does not reproduce their contents.

For a first real use, start from a bundled workflow rather than from an empty canvas. The 12/07 entry links a workflow file under workflows/ named Krea 2 Edit (Full Context).json, and the 29/06 entry links a Krea 2 image edit workflow. Load one of those JSON files through ComfyUI's workflow loader and check that the nodes it references resolve to entries in the Rebalance Pack menu. If the graph loads with missing-node placeholders, the pack did not import, and the usual cause is a clone placed outside custom_nodes or a Python import error printed in the ComfyUI console at startup. The README does not list dependencies, so any missing import has to be read from that console output.

Where Rebalance Pack stops being the right tool

The clearest limitation is documentation depth. The README is a dated changelog, and the pyproject.toml description is a one-line inventory. Neither states what inputs each node takes, what it returns, or which values are valid. For the model-specific nodes this matters most, because names like Conditioning Krea 2 Rebalance imply a specific conditioning transformation that the README never explains. You are expected to infer behaviour from the source files or from the linked workflows.

Versioning is the second constraint. pyproject.toml declares version 2.1.0, and the README's newest entry is dated 10/08, but there is no changelog section that maps versions to the dated entries, so you cannot tell which release introduced which node. The repository has no retrieved releases, which means there are no tagged artifacts to pin against. If your team needs reproducible node versions across machines, cloning main gives you whatever the branch holds at that moment.

The third case is scope. If your ComfyUI work does not involve Krea 2 or Ideogram 4, the model-specific half of the pack is dead weight, and the generic half competes with node packs that have per-node reference pages. There is also no stated support for ComfyUI Manager installation, no dependency list, and no documented rollback procedure if a new commit changes a node's behaviour under an existing workflow.

Maintenance status: the repository is not archived, and the last push was on 2026-08-12.

How it differs from a general-purpose ComfyUI utility pack

The natural alternative is a broad ComfyUI utility collection such as ComfyUI Essentials, which offers a wide catalogue of small, individually documented nodes for image, mask, and latent manipulation. The difference is in the design centre. A pack like Essentials is built around many fixed, narrow nodes, each with a stable signature that a workflow can depend on. Rebalance Pack is built around fewer nodes that do more, with Omni Node going furthest by deriving its own signature from pasted code.

That changes the failure mode. With fixed nodes, a workflow that loads is a workflow whose sockets exist. With a self-rebuilding node, a workflow can load and still be wrong, because the pasted class body is data rather than schema. The second difference is model coupling: Essentials is model-agnostic, while Rebalance Pack ships nodes whose names encode Krea 2 and Ideogram 4 conditioning and encoding paths, plus example workflows for those models. If your pipeline is those models, the coupling is the point. If it is not, the coupling is overhead.

A third option is writing your own custom node file for the one transformation you need. That is more work up front and gives you a signature you control, which is what the Omni Node approach is trying to avoid.

Licence and the cost of staying current

The repository is licensed Apache-2.0. That permits commercial use, modification, and redistribution provided you keep the licence and notice files and state significant changes. One detail to check before redistributing: pyproject.toml declares the licence as a file reference to LICENSE.txt, while the top-level repository listing shows a file named LICENSE. If you vendor the pack or ship it inside a product, confirm which file the metadata points to and include the actual licence text rather than assuming the two names refer to the same file. This is a packaging observation, not legal advice; consult your own counsel for compliance questions.

Upgrade cost is shaped by the same gap noted earlier. There are no tagged releases, so upgrading means pulling main and re-reading the dated README entries to see what moved. The dated entries are the only changelog the project publishes. If a workflow depends on a model-specific node whose behaviour changed in a commit, the README's newest entry is where you would notice, and only if that entry describes the change. Budget for reading the diff, not for reading release notes.

Editorial conclusion

Adopt it if you already run ComfyUI with the Krea 2 or Ideogram 4 model families and you want the model-specific conditioning and masking nodes in one pack, accepting that the README is a chronological changelog rather than a manual. Do not adopt it if you need a documented, versioned API, if your workflows do not touch those two model families, or if you cannot read Python well enough to check what a node does before wiring it in. Verify first that the node names in your ComfyUI menu match the ones listed in the README, that the bundled workflows under workflows/ load against your installed model files, and that pyproject.toml still declares version 2.1.0 and the Apache-2.0 licence file.

Frequently asked questions

Official sources

1. Issues
2. License: Apache-2.0
3. nova452/Rebalance-Pack on GitHub
4. README