[PR #3340] [CLOSED] Add ComfyUI-LiveChatStream (ketle-man) #3209

New issue

Closed

opened 2026-10-28 21:16:06 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-28 21:16:06 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/ComfyUI-Manager/pull/3340
Author: @ketle-man
Created: 10/3/2026
Status: ❌ Closed

Base: main ← Head: add-comfyui-livechatstream

📝 Commits (2)

c164ee5 Add ComfyUI-LiveChatStream (ketle-man)
f80716b Update ComfyUI-LiveChatStream description (decision model wording)

📊 Changes

2 files changed (+19 additions, -0 deletions)

View changed files

📝 custom-node-list.json (+11 -0)
📝 extension-node-map.json (+8 -0)

📄 Description

Adds ComfyUI-LiveChatStream to the custom node list.

Repository: https://github.com/ketle-man/ComfyUI-LiveChatStream
License: Apache-2.0
Node: LiveChatStream ("Live Chat Stream")

What it does

Chat with an LLM / VLM over a local Ollama server inside a node and turn the streaming reply into images: generation starts as soon as the prompt block is complete, while the rest of the reply is still streaming.

Two prompt lanes: a clean <prompt> block (lane A) and the model's own thinking / reply / inner monologue used as-is (lane B)
Character presets (stored server-side), image input / output for I2I, VLM attachments, optional decision-model image judge
Optional VRAM management (unload Ollama models before image generation)
UI localized for en / ja / zh via locales/<lang>/nodeDefs.json and the ComfyUI locale setting (no browser-language auto-binding)

Notes for review

No extra Python dependencies (dependencies = []); the node only talks to the user-configured Ollama host.
The server routes only accept loopback / private-network hosts for Ollama (resolved and checked server-side, redirects are not followed).
Free VRAM is read with nvidia-smi through subprocess.run with an argument list (no shell), falling back to ComfyUI's own value when it is unavailable.
Node description and tooltips are in English; translations are provided through locales/.

Changes

custom-node-list.json: new entry appended at the end
extension-node-map.json: new entry appended at the end

_{🔄 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/3340 **Author:** [@ketle-man](https://github.com/ketle-man) **Created:** 10/3/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `add-comfyui-livechatstream` --- ### 📝 Commits (2) - [`c164ee5`](https://github.com/Comfy-Org/ComfyUI-Manager/commit/c164ee5f844c365ff5da1bb4a37aee1b913c8897) Add ComfyUI-LiveChatStream (ketle-man) - [`f80716b`](https://github.com/Comfy-Org/ComfyUI-Manager/commit/f80716b5f8b0852400b757e8b7a609cd6a78f290) Update ComfyUI-LiveChatStream description (decision model wording) ### 📊 Changes **2 files changed** (+19 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `custom-node-list.json` (+11 -0) 📝 `extension-node-map.json` (+8 -0) </details> ### 📄 Description Adds **ComfyUI-LiveChatStream** to the custom node list. - Repository: https://github.com/ketle-man/ComfyUI-LiveChatStream - License: Apache-2.0 - Node: `LiveChatStream` ("Live Chat Stream") ### What it does Chat with an LLM / VLM over a local [Ollama](https://ollama.com) server inside a node and turn the **streaming** reply into images: generation starts as soon as the prompt block is complete, while the rest of the reply is still streaming. - Two prompt lanes: a clean `<prompt>` block (lane A) and the model's own thinking / reply / inner monologue used as-is (lane B) - Character presets (stored server-side), image input / output for I2I, VLM attachments, optional decision-model image judge - Optional VRAM management (unload Ollama models before image generation) - UI localized for en / ja / zh via `locales/<lang>/nodeDefs.json` and the ComfyUI locale setting (no browser-language auto-binding) ### Notes for review - No extra Python dependencies (`dependencies = []`); the node only talks to the user-configured Ollama host. - The server routes only accept loopback / private-network hosts for Ollama (resolved and checked server-side, redirects are not followed). - Free VRAM is read with `nvidia-smi` through `subprocess.run` with an argument list (no shell), falling back to ComfyUI's own value when it is unavailable. - Node description and tooltips are in English; translations are provided through `locales/`. ### Changes - `custom-node-list.json`: new entry appended at the end - `extension-node-map.json`: new entry appended at the end --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-28 21:16:06 +00:00