TerraVis: Towards Evaluation of World-Grounded Visual Consistency in Text-to-Image Generation via MLLM Workflows
TerraVis uses multimodal LLMs to score world-consistency failures in text-to-image outputs across 18 violation types.
1 stories, newest first · one card per story
TerraVis uses multimodal LLMs to score world-consistency failures in text-to-image outputs across 18 violation types.
Today 1.6K readers · 2.5K views · 780 reads · all time 189K views