Research map / Image generation / Autoregressive and latent generation
TWIGMA: A dataset of AI-Generated Images with Metadata From Twitter
Yiqun T. Chen, James Zou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Compare images less similar to ArtBench neighbors with images most similar to ArtBench neighbors | TWIGMA images, adjusted for follower category · TWIGMA; ArtBench similarity quintiles | likes: 19% more likes | Improved |
| Compare images in ArtBench similarity quintile 2 with images most similar to ArtBench neighbors | TWIGMA images, adjusted for follower category · TWIGMA; ArtBench similarity quintiles | likes: 12% more likes | Improved |
| Compare images in ArtBench similarity quintile 3 with images most similar to ArtBench neighbors | TWIGMA images, adjusted for follower category · TWIGMA; ArtBench similarity quintiles | likes: 0.8% fewer likes | Hurt |
| Compare images in ArtBench similarity quintile 4 with images most similar to ArtBench neighbors | TWIGMA images, adjusted for follower category · TWIGMA; ArtBench similarity quintiles | likes: 2% more likes | Improved |
| Compare images less similar to LAION neighbors with images most similar to LAION neighbors | TWIGMA images, adjusted for follower category · TWIGMA; LAION similarity quintiles | likes: 10% (95% CI: [9%, 10%]) | Improved |
| Compare images in LAION similarity quintile 2 with images most similar to LAION neighbors | TWIGMA images, adjusted for follower category · TWIGMA; LAION similarity quintiles | likes: 2% more (95% CI: [2%, 3%]) | Improved |
| Condition image pairs on identical prompts rather than different prompts | DiffusionDB · DiffusionDB | pairwise image-embedding distance: roughly half | Hurt |
| Use longer, more detailed prompts | DiffusionDB image generation · DiffusionDB | within-prompt output variation: Kendall's τ: -0.1, p-values: 0.01 | Hurt |
| Compare AI-generated image distributions with non-AI image distributions | TWIGMA and DiffusionDB · TWIGMA, DiffusionDB, LAION, and ArtBench | distributional distinctiveness | Improved |
| Compare image variability across datasets | TWIGMA · TWIGMA, LAION, ArtBench, and DiffusionDB | image variation | Hurt |
| Compare TWIGMA themes over time | TWIGMA · Twitter images from 2021 to 2023 | prevalence of artistically sophisticated themes | Improved |
| Compare TWIGMA themes over time | TWIGMA · Twitter images from 2021 to 2023 | prevalence of natural scenes | Hurt |
Closest papers by meaning
- GenAI Arena: An Open Evaluation Platform for Generative Models (NeurIPS 2024 · similarity 0.87)
- Social Reward: Evaluating and Enhancing Generative AI through Million-User Feedback from an Online Creative Community (ICLR 2024 · similarity 0.86)
- Stylebreeder: Exploring and Democratizing Artistic Styles through Text-to-Image Models (NeurIPS 2024 · similarity 0.86)
- ArtWhisperer: A Dataset for Characterizing Human-AI Interactions in Artistic Creations (ICML 2024 · similarity 0.86)
- Gender Artifacts from Art History to Text-to-Image Generation (NeurIPS 2026 · similarity 0.86)
- UnlearnCanvas: Stylized Image Dataset for Enhanced Machine Unlearning Evaluation in Diffusion Models (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
