[PR #2683] [MERGED] Add ComfyUI Qwen-VL LoRA #2690

New issue

Closed

opened 2026-10-28 21:11:49 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-28 21:11:49 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/Comfy-Org/ComfyUI-Manager/pull/2683
Author: @Dangocan
Created: 3/10/2026
Status: ✅ Merged
Merged: 3/15/2026
Merged by: @ltdrdata

Base: main ← Head: add-qwenvl-lora-node

📝 Commits (1)

9b2551f Add ComfyUI Qwen-VL LoRA node by Dangocan

📊 Changes

1 file changed (+12 additions, -0 deletions)

View changed files

📝 custom-node-list.json (+12 -0)

📄 Description

Add ComfyUI Qwen-VL LoRA to custom-node-list.json

Repository: https://github.com/Dangocan/comfyui_qwenvl_lora

What this node pack does

Three nodes that extend Qwen-VL / Qwen3-VL captioning workflows with LoRA adapter support:

Qwen-VL Model Loader — loads local Qwen-VL or Qwen3-VL checkpoints with selectable quantization (4-bit NF4, 8-bit, FP16), attention backend (auto / SageAttention / Flash Attention 2 / SDPA), device selection, FP8 pre-quantized model support, and optional torch.compile
Qwen-VL LoRA Loader — applies a PEFT LoRA adapter on top of any loaded model with an adjustable strength slider (0–2)
Qwen-VL Caption — runs inference on a ComfyUI IMAGE tensor and outputs a text caption with a fully configurable prompt

JSON entry added

{
    "author": "Dangocan",
    "title": "ComfyUI Qwen-VL LoRA",
    "id": "comfyui-qwenvl-lora",
    "reference": "https://github.com/Dangocan/comfyui_qwenvl_lora",
    "files": [
        "https://github.com/Dangocan/comfyui_qwenvl_lora"
    ],
    "install_type": "git-clone",
    "description": "Load Qwen-VL and Qwen3-VL models locally and apply PEFT LoRA adapters for enhanced image captioning. Supports 4-bit and 8-bit quantization (BitsAndBytes), FP8 pre-quantized models, Flash Attention 2, SageAttention, and torch.compile. Includes a LoRA strength slider and configurable captioning prompts.",
    "tags": ["captioning", "LLM", "VLM", "LoRA", "Qwen"]
}

Checklist

Only custom-node-list.json is modified
JSON validated locally (python -c "import json; json.load(open('custom-node-list.json'))")
All six required fields present: author, title, reference, files, install_type, description
files URL does not end in .git
id field included (lowercase hyphenated)
Description is user-facing (not a list of install requirements)
Entry inserted adjacent to existing Qwen-VL nodes (after ComfyUI-QwenVL by 1038lab)
Node pack is complementary to ComfyUI-QwenVL by 1038lab — it adds LoRA adapter loading on top of their model type, not a replacement

Credits

Loading internals (quantization config, FP8 path, attention mode resolution, SageAttention patching) are derived from ComfyUI-QwenVL by 1038lab (GPL-3.0). The repository is also licensed GPL-3.0.

_{🔄 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/2683 **Author:** [@Dangocan](https://github.com/Dangocan) **Created:** 3/10/2026 **Status:** ✅ Merged **Merged:** 3/15/2026 **Merged by:** [@ltdrdata](https://github.com/ltdrdata) **Base:** `main` ← **Head:** `add-qwenvl-lora-node` --- ### 📝 Commits (1) - [`9b2551f`](https://github.com/Comfy-Org/ComfyUI-Manager/commit/9b2551ffc83242c87d0c1a4ece384d4218d60193) Add ComfyUI Qwen-VL LoRA node by Dangocan ### 📊 Changes **1 file changed** (+12 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `custom-node-list.json` (+12 -0) </details> ### 📄 Description ## Add ComfyUI Qwen-VL LoRA to custom-node-list.json **Repository:** https://github.com/Dangocan/comfyui_qwenvl_lora ### What this node pack does Three nodes that extend Qwen-VL / Qwen3-VL captioning workflows with LoRA adapter support: - **Qwen-VL Model Loader** — loads local Qwen-VL or Qwen3-VL checkpoints with selectable quantization (4-bit NF4, 8-bit, FP16), attention backend (auto / SageAttention / Flash Attention 2 / SDPA), device selection, FP8 pre-quantized model support, and optional `torch.compile` - **Qwen-VL LoRA Loader** — applies a PEFT LoRA adapter on top of any loaded model with an adjustable strength slider (0–2) - **Qwen-VL Caption** — runs inference on a ComfyUI `IMAGE` tensor and outputs a text caption with a fully configurable prompt ### JSON entry added ```json { "author": "Dangocan", "title": "ComfyUI Qwen-VL LoRA", "id": "comfyui-qwenvl-lora", "reference": "https://github.com/Dangocan/comfyui_qwenvl_lora", "files": [ "https://github.com/Dangocan/comfyui_qwenvl_lora" ], "install_type": "git-clone", "description": "Load Qwen-VL and Qwen3-VL models locally and apply PEFT LoRA adapters for enhanced image captioning. Supports 4-bit and 8-bit quantization (BitsAndBytes), FP8 pre-quantized models, Flash Attention 2, SageAttention, and torch.compile. Includes a LoRA strength slider and configurable captioning prompts.", "tags": ["captioning", "LLM", "VLM", "LoRA", "Qwen"] } ``` ### Checklist - [x] Only `custom-node-list.json` is modified - [x] JSON validated locally (`python -c "import json; json.load(open('custom-node-list.json'))"`) - [x] All six required fields present: `author`, `title`, `reference`, `files`, `install_type`, `description` - [x] `files` URL does not end in `.git` - [x] `id` field included (lowercase hyphenated) - [x] Description is user-facing (not a list of install requirements) - [x] Entry inserted adjacent to existing Qwen-VL nodes (after `ComfyUI-QwenVL` by 1038lab) - [x] Node pack is complementary to `ComfyUI-QwenVL` by 1038lab — it adds LoRA adapter loading on top of their model type, not a replacement ### Credits Loading internals (quantization config, FP8 path, attention mode resolution, SageAttention patching) are derived from [ComfyUI-QwenVL](https://github.com/1038lab/ComfyUI-QwenVL) by 1038lab (GPL-3.0). The repository is also licensed GPL-3.0. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-28 21:11:49 +00:00