Downloads · 30 days
9.2K
17% of all-time downloads
Qwen2D-VAE
Anzhc/Qwen2D-VAE
Qwen2D-VAE is a machine learning model from Anzhc. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for diffusers. The card lists the license as apache-2.0.
A modification of Qwen-Image-VAE to remove temporal dimension, collapsing it from 3D to 2D. Primarily meant for caching latents for training, but also can be used in ComfyUI to save a second ever so often.
All-time downloads
54.7K
Public
Repo size
178 MB
Likes
28
Public
Repo makeup
Click a slice to open those files.
.safetensors178 MB · 100%
At a glance
- Library
- diffusers
- License
- apache-2.0
- Access
- Public
- Created
- Apr 17, 2026
- Updated
- May 22, 2026
- SHA
- 35ef9709
Base models
Model card
From the Hugging Face model README
A modification of Qwen-Image-VAE to remove temporal dimension, collapsing it from 3D to 2D. Primarily meant for caching latents for training, but also can be used in ComfyUI to save a second ever so often.
Why?
Because you don't need temporal dimension for Image models (unless you're doing research related to temporally consistent gens on them).
Collapsing it to 2D does not alter output, while reducing required VRAM 3x, and speeding it up 2.5x.
But what if im using temporal dimension for control input?
You don't encode them in temporal dimension, they are separate latents. It works.
Fully compatible with image models using Qwen/Wan VAEs.
Use this node pack to use it in your ComfyUI - https://github.com/Anzhc/anzhc-qwen2d-comfyui
- Library
- diffusers
- License
- apache-2.0
- Created
- Apr 17, 2026
- Updated
- May 22, 2026