imagebench: 50-Prompt Open-Source Benchmark Testing 8 Image Models on a DGX Spark
Aware_Web9715 · reddit · 2026-09-05
Redditor mazinb released imagebench, an open-source benchmark of 50 prompts for evaluating open-source image generation models, after failing to find a good comparison method. They ran 8 models on a DGX Spark and published all outputs and stats in a reference gallery. Useful for model selection and building your own image-model evals; the author is soliciting better evaluation approaches.
Related event: Developer Open-Sources imagebench to Benchmark 8 Text-to-Image Models(2 posts)→
More from Multimodal
- Kling AI officially launches Kling 4.0: 15 references per task, 30s clips, 10 keyframes — testingcatalog · 2026-10-23
- Kling 4.0 hands-on: multi-shot cyberpunk robot demo shows strong character consistency — testingcatalog · 2026-10-23
- Run Lola Run director Tom Tykwer uses Seedance in hybrid AI filmmaking experiment — Uncanny_Harry · 2026-10-23
- "Still Worth Something" S01E01: a Pixar-style fully AI-animated comedy short — LPGroves · 2026-10-23
- Music publisher Reservoir sues Suno and Anthropic over AI music training — TinfoilTricorn · 2026-10-23
- Halloween AI short: "Careful who you bring home" — glagolef · 2026-10-23