[PR #3376] [MERGED] Add ComfyUI-MacOffload-TE to custom node list #3243

New issue

Closed

opened 2026-10-28 21:16:24 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-28 21:16:24 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/ComfyUI-Manager/pull/3376
Author: @j0eyhu
Created: 10/14/2026
Status: ✅ Merged
Merged: 10/18/2026
Merged by: @ltdrdata

Base: main ← Head: patch-1

📝 Commits (1)

a2fc3e1 Add ComfyUI-MacOffload-TE to custom node list

📊 Changes

1 file changed (+10 additions, -0 deletions)

View changed files

📝 custom-node-list.json (+10 -0)

📄 Description

Plugin Overview

ComfyUI-MacOffload-TE eliminates VRAM thrashing on consumer GPUs by offloading heavy text encoders (e.g., Flux, Qwen-Image) to a secondary macOS / Apple Silicon device over local network streaming.

Key Features

0 MB Local VRAM Footprint: Text encoders remain fully resident in macOS Unified Memory (MPS), saving 8GB–18GB of local GPU memory.
Zero PCIe Thrashing: Eliminates model swapping between encoding and diffusion sampling, allowing the local GPU to remain primed with diffusion weights for immediate sampling.
Higher Native Resolutions: Reclaims headroom for DiT and VAE decoding on 8GB consumer GPUs (e.g., RTX 4060), enabling larger native latents without OOM crashes.
Client/Server Architecture:
- Client (Windows/Linux): Custom ComfyUI nodes running sampling and VAE decode.
- Server (macOS / Apple Silicon): Lightweight backend handling tokenization and forward passes via low-latency tensor streaming.

Repository

https://github.com/j0eyhu/ComfyUI-MacOffload-TE

_{🔄 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/3376 **Author:** [@j0eyhu](https://github.com/j0eyhu) **Created:** 10/14/2026 **Status:** ✅ Merged **Merged:** 10/18/2026 **Merged by:** [@ltdrdata](https://github.com/ltdrdata) **Base:** `main` ← **Head:** `patch-1` --- ### 📝 Commits (1) - [`a2fc3e1`](https://github.com/Comfy-Org/ComfyUI-Manager/commit/a2fc3e1c2103b6c1557812a77b45ac1d2b8699c0) Add ComfyUI-MacOffload-TE to custom node list ### 📊 Changes **1 file changed** (+10 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `custom-node-list.json` (+10 -0) </details> ### 📄 Description ### Plugin Overview **ComfyUI-MacOffload-TE** eliminates VRAM thrashing on consumer GPUs by offloading heavy text encoders (e.g., Flux, Qwen-Image) to a secondary macOS / Apple Silicon device over local network streaming. ### Key Features - **0 MB Local VRAM Footprint**: Text encoders remain fully resident in macOS Unified Memory (MPS), saving 8GB–18GB of local GPU memory. - **Zero PCIe Thrashing**: Eliminates model swapping between encoding and diffusion sampling, allowing the local GPU to remain primed with diffusion weights for immediate sampling. - **Higher Native Resolutions**: Reclaims headroom for DiT and VAE decoding on 8GB consumer GPUs (e.g., RTX 4060), enabling larger native latents without OOM crashes. - **Client/Server Architecture**: - **Client (Windows/Linux)**: Custom ComfyUI nodes running sampling and VAE decode. - **Server (macOS / Apple Silicon)**: Lightweight backend handling tokenization and forward passes via low-latency tensor streaming. ### Repository https://github.com/j0eyhu/ComfyUI-MacOffload-TE --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-28 21:16:24 +00:00