Qwen3-VL for ComfyUI

Qwen3-VL for ComfyUI

Repackaged Qwen3-VL 4B and 8B model files for ComfyUI's text encoder folder, offered in several precisions from bf16 to 4-bit styles. For ComfyUI users who need a Qwen3-VL text encoder.

Made by Comfy-Org on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: These are Qwen3-VL model files repackaged for ComfyUI. They come in 4B and 8B sizes and in several precisions, and they belong in the text_encoders folder. The card is short and does not say which workflows use them. The originals are the Qwen3-VL 4B Instruct and 8B Instruct models from Qwen.

Who it suits: It suits ComfyUI users whose workflow asks for a Qwen3-VL text encoder and who want a ready file in the right format.

Industry | Who uses it | How they use it | Example
Design and art | ComfyUI artists | Load a Qwen3-VL text encoder for a workflow that needs one | They pick the 4B file in the encoder node
Software development | Developers who build ComfyUI pipelines | Swap between sizes and precisions in a workflow | They try the 8B file in place of the 4B file
Local AI hobby | Users with limited memory | Choose a smaller precision of the same model | They try an fp8 or 4-bit style file

You can skip it if you do not use ComfyUI, or if your workflow does not call for a Qwen3-VL text encoder.

Where to use it:

• ComfyUI, on Windows, macOS or Linux
• The text_encoders folder under ComfyUI/models
• Python with the Hugging Face tools, to download the files

Use case | Who | What they get back
Provide a text encoder | ComfyUI users | A Qwen3-VL file the workflow can load
Try a smaller file | Users with limited memory | A lower-precision copy of the same model
Compare sizes | Pipeline developers | 4B and 8B files side by side

How to set it up: You need ComfyUI and a place for the downloaded files.

• A working ComfyUI install
• Python with the huggingface_hub package
• Disk space for the files you choose to keep

1. Install the Hugging Face tools with Python.
2. Download the repository files with the second command below.
3. Place the files you want in the text_encoders folder under ComfyUI/models.
4. Restart ComfyUI if it was open, then pick the file in your workflow.

pip install -U huggingface_hub
hf download Comfy-Org/Qwen3-VL

Title: The seven files, all for the text_encoders folder
Model size | Precision | File
4B | bf16 | qwen3vl_4b_bf16.safetensors
4B | fp8 scaled | qwen3vl_4b_fp8_scaled.safetensors
8B | bf16 | qwen3vl_8b_bf16.safetensors
8B | fp8 scaled | qwen3vl_8b_fp8_scaled.safetensors
8B | int8 convrot | qwen3vl_8b_int8_convrot.safetensors
8B | w4a8 | qwen3vl_8b_w4a8.safetensors
8B | nvfp4 | qwen3vl_8b_nvfp4.safetensors

How to use it: The card gives no command and names no workflow. After the files are in place, choose one in a ComfyUI node that loads a text encoder, and check that workflow's own notes for the file it expects.

Tool | What it does
text_encoders folder | Where ComfyUI looks for these Qwen3-VL files
Hugging Face download command | Fetches every file in the repository

Example: You place the 4B fp8 scaled file in the text_encoders folder and select it in a workflow that asks for a text encoder. The workflow then reads your prompt through that encoder. The card does not describe the workflow itself.

For newcomers and experts: Think of these as the language-reading part that some ComfyUI workflows need. A first small try is to download one file, place it in the folder and select it in a workflow that asks for it.

The file names show two model sizes and five precision styles: bf16, fp8 scaled, int8 convrot, w4a8 and nvfp4. The card does not compare them, so check your workflow's notes before choosing. Smaller precisions usually trade some accuracy for less memory, but the card gives no measurements.

Download the repository files
Place one in ComfyUI/models/text_encoders
Select it in a text encoder node
Run the workflow that uses it

Good to know:

• Repackaged files: they are Qwen3-VL weights arranged for ComfyUI, and the originals are Qwen's 4B Instruct and 8B Instruct models
• Short card: there is no workflow list, so check each workflow's own notes
• Other versions: other people have published 1 quantized version
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download and the prompts you run in ComfyUI
• Not for: use outside ComfyUI, since the files are repackaged for it

On Hugging Face, 13 November 2026

• Library: diffusion-single-file
• Downloads in the last 30 days: 1,223,409
• Likes: 118
• License: apache-2.0
• Based on: Qwen/Qwen3-VL-4B-Instruct
• Quantized versions: 1
• Tags: diffusion-single-file, comfyui

Links

• Model: https://huggingface.co/Comfy-Org/Qwen3-VL
• Files: https://huggingface.co/Comfy-Org/Qwen3-VL/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Comfy-Org/Qwen3-VL

Then follow the usage in its model card. Nothing in this kit runs by itself.