, impossible spatial relationships).
New metric detects real-world plausibility failures in AI-generated images
TerraVis uses a taxonomy of violations and a multi-stage MLLM workflow to score world-grounded consistency.
Big Tech
Shuai Fu · Jing Gu · Jian Zhou · Zicheng Duan · Gengze Zhou · Qi Wu
Adelaide University · xAI
Research Digest··2 min read
The authors introduce world-grounded visual consistency as a new evaluation dimension for text-to-image models.
Why this paper
From xAI and Adelaide University
In one line
TerraVis evaluates world-grounded visual consistency in generated images by detecting real-world plausibility violations using an MLLM workflow.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§