Skip to content

VQAScore

An evaluation metric for text-to-visual generation

Visit official site

What VQAScore is

The metric is described as a way to evaluate complex text-to-visual generation, introduced together with the GenAI-Bench benchmark set.

What you can do with it

  • Score text-to-image generation results
  • Compare models on the GenAI-Bench prompts
  • Evaluate text-to-video and 3D generation

Model notes

  • The CLIP-FlanT5 VQA model sits behind the score
  • The approach is presented as an alternative to CLIPScore
  • GenAI-Bench provides real-scenario test texts

Pros & cons

✓ What we like

  • GenAI-Bench supplies prompts with rich compositional semantics
  • The metric covers image, video and 3D generation

! What to watch out for

  • The score is built on a specific VQA model
  • Evaluation results depend on the benchmark prompts

FAQ

What model is it based on?

The material states VQAScore is based on CLIP-FlanT5.

What is GenAI-Bench?

The material describes a benchmark set of test texts with compositional semantics.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review