CogVLM 2 Vision Model App: 1-Click Batch Image Captioning Installers
CogVLM 2, the most powerful open source vision model at GPT-4 level for image tasks, in a SECourses batch captioning app that runs fast in 4-bit mode.

Highlights
1-click installers for CogVLM V2 on Windows, RunPod and Massed Compute with Torch 2.7, CUDA 12.8 and RTX 5000 support. FP16 needs a 48 GB GPU, and the V10 update fixed image orientation and multi-GPU batch processing.
What it does
CogVLM 2 is an open source vision model that the post calls GPT-4 level for vision tasks. This Gradio app, developed by SECourses, captions images one by one or whole folders in batch and runs very fast with 4-bit quantization. The example captions were made locally on Windows on an RTX 3090 in 4-bit mode, and FP16 works on 48 GB GPUs.
What you get
- CogVLM2_v10 zip with 1-click installers for Windows, RunPod and Massed Compute
- Torch 2.7 and CUDA 12.8 with RTX 5000 series support, plus pre-compiled xFormers, Flash Attention, Sage Attention (Linux) and Triton
- Batch captioning with multi-GPU support and correct image orientation (V10, 1 May 2025)
- On Windows, the models download into the app's models folder (around 37 GB)
Requirements
- Windows: Python 3.10, FFmpeg, CUDA 12.4, cuDNN, C++ tools, MSVC and Git
- Follow the Windows requirements tutorial
- If it runs too slowly, your VRAM is likely not enough: use 4-bit mode
- Cloud setup: Massed Compute and RunPod tutorial
Official repository: THUDM/CogVLM2
Video tutorial
Set Up Windows 11 for AI Apps: Python, CUDA, cuDNN, C++ Build Tools, FFmpeg and Git (2024 Guide) · chapters and summary
More videos: Native 4 MP AI Images, Fast on Low VRAM: NVIDIA SANA Step-by-Step Tutorial (Windows, Cloud, Kaggle)
Related

All JoyCaption Apps: 1-Click Installers With Batch Captioning
JoyCaption in four improved Gradio apps: Pre-Alpha, Alpha One, Alpha Two and the newest, fastest Joy Caption Beta One, all with batch and multi-GPU captioning.

Ultimate Image Processing App: 1-Click Batch Cropping, SAM 3.1 Masking
Ultimate Image Processing Premium turns raw photo folders into training-ready datasets with nine tools, from SAM 3.1 cropping to duplicate removal.

SECourses Caption Editor Pro: Easy Batch Caption Editing Gradio App
SECourses Caption Editor Pro edits image captions for training datasets one by one or in batch, with tag analytics, a token counter and a check-word tab.