Holistic Evaluation of Text-To-Image Models
by Stanford CRFM · free · Last verified 2026-03-17
Presents HEIM, a comprehensive framework for evaluating text-to-image models across 12 aspects like alignment, quality, aesthetics, bias, and toxicity. The study benchmarks 26 models, revealing that no single model excels in all areas and highlighting significant safety gaps in current generative AI.
https://arxiv.org/abs/2311.04287 ↗B
B—Above Average
Adoption: BQuality: AFreshness: BCitations: BEngagement: F
Specifications
- License
- Apache-2.0
- Pricing
- free
- Capabilities
- text-to-image model evaluation, multi-aspect performance assessment, social bias and fairness auditing, toxicity and safety analysis, image quality and aesthetics scoring, originality and compositionality testing, reasoning and knowledge evaluation, comparative model benchmarking
- Integrations
- Use Cases
- [object Object], [object Object], [object Object], [object Object]
- API Available
- No
- Tags
- evaluation, text-to-image, holistic-evaluation, benchmark, multimodal-ai, ai-safety, generative-ai, responsible-ai, model-comparison, ai-ethics
- Added
- 2026-03-17
- Completeness
- 0.9%
Index Score
61.8Adoption
68
Quality
88
Freshness
68
Citations
68
Engagement
0