JLC ComfyUI Nodes

<p align="center"> <img src="assets/icons/jlc-comfyui-nodes_Logo-0512.png" width="120"> &nbsp;&nbsp;&nbsp; <img src="assets/icons/jlc-comfyui-nodes_Logo-Dark-0512.png" width="120"> </p> <p align="center"> <a href="https://registry.comfy.org/packages/jlc-comfyui-nodes"> <img src="https://img.shields.io/badge/Available%20on-ComfyUI%20Registry-blue" alt="ComfyUI Registry"> </a> <img src="https://img.shields.io/badge/ComfyUI-Custom%20Nodes-blue" alt="ComfyUI Custom Nodes"> <img src="https://img.shields.io/badge/license-MIT-green" alt="MIT License"> <img src="https://img.shields.io/badge/status-active-brightgreen" alt="Status: active"> </p>

[!IMPORTANT] ControlNet compatibility update — prior warning withdrawn

The temporary ControlNet compatibility warning previously shown here is no longer applicable. The current ControlNet Composition, Orchestrator, Orchestrator Advanced, Apply Advanced, shared composition core, and dynamic slot-visibility paths have been audited, corrected, and validated against current ComfyUI sampler and model-management interfaces.

Compared with the prior release, the revised implementation provides substantially improved VRAM handling and markedly reduced execution time in the tested Flux and multi-ControlNet workflows. These gains come from internal composition optimizations, corrected cache and lifecycle integration, removal of unnecessary synchronization from the normal path, and alignment with current ComfyUI architecture.

Validation included demanding Flux workflows with multiple LoRAs, Union ControlNet models, repeated ControlNet use, and as many as four ControlNet slots. The previously observed catastrophic slowdown was traced primarily to launching ComfyUI with forced --lowvram, not to the JLC non-recursive composition architecture.

The current implementation was validated against ComfyUI commit 2a610155 from June 22, 2026, with frontend package 1.45.19. See the detailed ControlNet guide for the full compatibility baseline, benchmark context, and runtime recommendations.

JLC ComfyUI Nodes is a custom-node collection built around Non-Recursive ControlNet Composition, a method introduced by this project to replace recursively nested multi-ControlNet evaluation with a flattened composition path whose execution cost scales approximately linearly with the number of applied ControlNets, rather than accumulating the severe repeated work, runtime growth, and memory pressure of native recursive chains.

The collection also includes supporting tools for padded inpainting and outpainting, dynamic ControlNet auxiliary preprocessing, dynamic LoRA loading, stage-boundary VRAM cleanup, and other practical workflow utilities.

Developed by J. L. Córdova, the project is especially focused on Flux-oriented image-generation pipelines, ControlNet-heavy workflows, multi-stage inference, LoRA experimentation, and advanced inpainting and outpainting.

---

Start Here: Release 2.0 Showcase Workflows

These showcase workflows demonstrate the breadth of the node pack and provide practical starting points for exploring Release 2.0.

The JLC ControlNet Orchestrator (Advanced) workflow serves both as an approachable starting point for new users and as a reference implementation of the package’s non-recursive ControlNet composition system.

Download PNG workflow · Download JSON workflow

The JLC All-In-One Workflow presents the collection as a complete working toolchain rather than a set of isolated nodes. It combines ControlNet with representative nodes from the ControlNet Aux, Dynamic LoRA Loader, padded-image and padded-latent, and utility families.

Download PNG workflow · Download JSON workflow

As with the other workflows included in this documentation, each example is provided both as a PNG with an embedded workflow for direct drag-and-drop into ComfyUI and as a JSON file for standard workflow loading.

---

Documentation

This README is the front page for the repository. Detailed documentation is organized by node family:

1. ControlNet Composition and Orchestration
2. Padded Image / Padded Latent
3. ControlNet Aux Preprocessor Wrappers
4. Dynamic LoRA Loaders
5. Utility Nodes

---

Node Families

1. ControlNet Composition and Orchestration

A family of nodes for native ControlNet application, modular chain construction, and explicit linearized non-recursive weighted fusion.

This family includes:

• JLC ControlNet Composition
• JLC ControlNet Orchestrator
• JLC ControlNet Orchestrator (Advanced)
• JLC ControlNet Apply
• JLC ControlNet Apply (Advanced)

The central design treats prepared ControlNets as independent operators evaluated against the same sampler state and combines their outputs through explicit weighted addition. The modular Apply Advanced → Composition workflow and the integrated Orchestrator Advanced workflow use the same validated fusion core.

• Orchestrator Advanced is the recommended integrated interface for most new multi-ControlNet workflows.
• Apply Advanced → Composition is a first-class modular interface when explicit chaining, pass-through wiring, or separate model sourcing is useful.
• Orchestrator remains the specialized external-input interface for ControlNet objects supplied by standard, third-party, custom, or nonstandard-location loaders.

Read the ControlNet guide

---

2. Padded Image / Padded Latent

Nodes for padded-canvas workflows, especially inpainting and outpainting.

This family includes:

• JLC Padded Image
• JLC Inpaint-Conditioned Padded Latent

Use Padded Image when you want image, mask, and canvas preparation while keeping VAE encoding and inpaint conditioning separate. Use Padded Latent when you want a more integrated node that prepares the padded canvas and also injects inpaint-conditioning metadata.

Read the Padded Image / Padded Latent guide

---

3. ControlNet Aux Preprocessor Wrappers

A dynamic convenience wrapper for simple image-in/image-out preprocessors provided by Fannovel16's comfyui_controlnet_aux package.

This family currently includes:

• JLC Dynamic Aux Preprocessor Wrapper

The wrapper does not replace Fannovel16's nodes and does not claim ownership of the underlying preprocessors. It provides a compact multi-slot interface for JLC workflows that need several simple ControlNet hint images from the same source image. Parameter-heavy preprocessors should still be used through their native ControlNet Aux nodes.

comfyui_controlnet_aux must be installed for non-disabled preprocessor slots to run.

Read the ControlNet Aux guide

---

4. Dynamic LoRA Loaders

Dynamic LoRA loader nodes for MODEL-only and MODEL+CLIP workflows, including shared and per-slot block-weight variants.

This family includes:

• JLC LoRA Loader - Multi Model
• JLC LoRA Loader - Multi-Model / CLIP
• JLC LoRA Loader - Multi-Model / Shared Block Weight
• JLC LoRA Loader - Multi Model / Shared Block Weight + CLIP
• JLC LoRA Loader - Multi-Model / Block Weight
• JLC LoRA Loader - Multi-Model / CLIP + Block Weight

These nodes predeclare up to ten LoRA slots and use frontend visibility controls to expose only the active rows. Hidden slot values remain serialized in workflow JSON, while the backend treats slot_count as authoritative. MODEL-only variants intentionally avoid CLIP/text-encoder patching; MODEL+CLIP variants expose independent MODEL and CLIP strengths.

Read the LoRA Loader guide

---

5. Utility Nodes

Small workflow-support nodes for seed discipline and stage-boundary memory hygiene.

This family includes:

• JLC Seed Generator — shared seed source that keeps the visible base seed stable while a frontend display reports the last seed actually used.
• JLC Stage Boundary VRAM Cleanup — experimental latent-passthrough cleanup helper for advanced multi-stage workflows where selected heavy model objects should be unloaded before the next stage.

Read the Utility Nodes guide

---

Installation

Install through the ComfyUI Registry:

https://registry.comfy.org/packages/jlc-comfyui-nodes

Manual install:

cd ComfyUI/custom_nodes
git clone https://github.com/Damkohler/jlc-comfyui-nodes.git

Restart ComfyUI after installation or update.

To update manually:

cd ComfyUI/custom_nodes/jlc-comfyui-nodes
git pull

Optional dependency: ControlNet Aux preprocessors

The JLC Dynamic Aux Preprocessor Wrapper requires Fannovel16's comfyui_controlnet_aux package when any non-disabled preprocessor slot is used.

Install it through ComfyUI Manager, or follow the upstream installation instructions:

https://github.com/Fannovel16/comfyui_controlnet_aux

Some upstream preprocessors may download or load large auxiliary models the first time they are used.

---

Compatibility Notes

The nodes are designed for ComfyUI custom-node workflows and have been developed primarily around:

• Flux-based image generation
• ControlNet-heavy pipelines
• LoRA experimentation
• inpainting and outpainting
• multi-stage inference graphs

The current ControlNet family is aligned with the ComfyUI sampler, ControlNet, model-management, hook, and lifecycle interfaces present at the tested baseline documented in the ControlNet guide. Later ComfyUI revisions may require renewed testing if those interfaces change.

For the validated 16 GB RTX 4090 Laptop workflow, use normal ComfyUI VRAM behavior with DynamicVRAM where desired. Forced --lowvram is not recommended for the benchmarked Flux and multi-ControlNet configurations because it caused destructive partial unload/reload cycles and severe execution-time regression.

Real MultiGPU ControlNet cloning is not implemented by the JLC composed wrapper. The compatibility attributes present in the wrapper are single-device shunts, not a claim of MultiGPU support.

---

Repository Structure

The current repository includes these main areas:

nodes/
  controlnet_aux_nodes/
  engines/
  lora_loader_nodes/
  util_nodes/
web/
assets/
  icons/
  workflows/
docs/

A generated repository map is also included:

jlc-comfyui-nodes-structure.txt

---

License

MIT License.

---

Author

J. L. Córdova
GitHub: Damkohler

---

Contributions

Suggestions, bug reports, and improvements are welcome through GitHub issues or pull requests.