CLIP ViT-H/14: Vision Encoder

CLIP ViT-H/14 trained on LAION-2B, a large vision encoder model repackaged for ComfyUI as a CLIP vision model for conditioning.

CLIP ViT-H/14 is a large vision encoder model trained on the LAION-2B dataset (32B samples seen, 79K steps). It serves as a CLIP vision encoder for conditioning in ComfyUI, providing high-quality image embeddings for various diffusion model conditioning tasks.

Usage in ComfyUI

Place the model file in ComfyUI/models/clip_vision/ (the split_files/clip_vision/ directory maps to this). Compatible with ComfyUI's standard CLIP vision loader node.

Developer	LAION / OpenCLIP
Release Date	2023 (repackaged 2025)
Architecture	ViT-H/14
License	MIT
Type	CLIP Vision Encoder
Training Data	LAION-2B