[PR #3073] [CLOSED] [Add Custom Node] ComfyUI-ModelMergeCosmosPredict-2B-Slerp #2994

New issue

Closed

opened 2026-10-28 21:14:22 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-28 21:14:22 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/ComfyUI-Manager/pull/3073
Author: @rikunarita
Created: 7/10/2026
Status: ❌ Closed

Base: main ← Head: patch-1

📝 Commits (1)

7800289 Update custom-node-list.json

📊 Changes

1 file changed (+7 additions, -54241 deletions)

View changed files

📝 custom-node-list.json (+7 -54241)

📄 Description

Summary

This PR adds a new custom node, ModelMergeCosmosPredict2_2B_Slerp, to the ComfyUI Manager custom node list.
This node is specifically designed for merging Cosmos Predict 2B models using Slerp (Spherical Linear Interpolation), while also supporting standard Linear interpolation.

Motivation

Currently, the official ModelMergeCosmosPredict2_2B node only supports Linear (weighted average) interpolation. When merging models with significantly different styles or characteristics, Linear interpolation can sometimes cause the output to become blurry or lose feature magnitude.
Slerp preserves the magnitude of the weight vectors, resulting in sharper and more stable merged models. This custom node fills that gap for Cosmos Predict 2B users.

Features

Slerp & Linear Support: Easily switch between Slerp and Linear interpolation via a dropdown menu.
Layer-wise Ratio Control: Provides individual sliders for all 28 Transformer blocks, as well as embedding and final layers, allowing precise layer-by-layer blending.
Numerical Stability: Includes strict safeguards (zero-norm prevention, dot-product clamping, and automatic fallback to Linear interpolation) to prevent NaN errors and crashes.
Drop-in Replacement: Fully compatible with the existing Cosmos Predict 2B workflow structure.

Node Details

Node Name: ModelMergeCosmosPredict2_2B_Slerp
Display Name: Model Merge Cosmos Predict 2B (Slerp)
Category: model/merging/model specific
Repository: https://github.com/rikunarita/ComfyUI-ModelMergeCosmosPredict-2B-Slerp

Checklist

The repository is public and accessible.
__init__.py and nodes.py are correctly structured and tested in a ComfyUI environment.
A comprehensive README.md is included in the repository.
No external dependencies are required (uses only standard ComfyUI/PyTorch libraries).

Thank you for reviewing and maintaining the ComfyUI Manager list! Please let me know if any changes are required.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/Comfy-Org/ComfyUI-Manager/pull/3073 **Author:** [@rikunarita](https://github.com/rikunarita) **Created:** 7/10/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `patch-1` --- ### 📝 Commits (1) - [`7800289`](https://github.com/Comfy-Org/ComfyUI-Manager/commit/7800289d7a34d3c9830e61095330e607e01b78eb) Update custom-node-list.json ### 📊 Changes **1 file changed** (+7 additions, -54241 deletions) <details> <summary>View changed files</summary> 📝 `custom-node-list.json` (+7 -54241) </details> ### 📄 Description ## Summary This PR adds a new custom node, `ModelMergeCosmosPredict2_2B_Slerp`, to the ComfyUI Manager custom node list. This node is specifically designed for merging **Cosmos Predict 2B** models using **Slerp (Spherical Linear Interpolation)**, while also supporting standard Linear interpolation. ## Motivation Currently, the official `ModelMergeCosmosPredict2_2B` node only supports Linear (weighted average) interpolation. When merging models with significantly different styles or characteristics, Linear interpolation can sometimes cause the output to become blurry or lose feature magnitude. Slerp preserves the magnitude of the weight vectors, resulting in sharper and more stable merged models. This custom node fills that gap for Cosmos Predict 2B users. ## Features - **Slerp & Linear Support**: Easily switch between Slerp and Linear interpolation via a dropdown menu. - **Layer-wise Ratio Control**: Provides individual sliders for all 28 Transformer blocks, as well as embedding and final layers, allowing precise layer-by-layer blending. - **Numerical Stability**: Includes strict safeguards (zero-norm prevention, dot-product clamping, and automatic fallback to Linear interpolation) to prevent NaN errors and crashes. - **Drop-in Replacement**: Fully compatible with the existing Cosmos Predict 2B workflow structure. ## Node Details - **Node Name**: `ModelMergeCosmosPredict2_2B_Slerp` - **Display Name**: Model Merge Cosmos Predict 2B (Slerp) - **Category**: `model/merging/model specific` - **Repository**: https://github.com/rikunarita/ComfyUI-ModelMergeCosmosPredict-2B-Slerp ## Checklist - [x] The repository is public and accessible. - [x] `__init__.py` and `nodes.py` are correctly structured and tested in a ComfyUI environment. - [x] A comprehensive `README.md` is included in the repository. - [x] No external dependencies are required (uses only standard ComfyUI/PyTorch libraries). Thank you for reviewing and maintaining the ComfyUI Manager list! Please let me know if any changes are required. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-28 21:14:22 +00:00