Adobe releases TPIPS: Text-Prompted Image Perceptual Similarity metric
junyanz89 · x · 2026-08-20
Adobe Research, in collaboration with CMU and UC Berkeley, has released TPIPS (Text-Prompted Image Perceptual Similarity). Unlike traditional metrics like FID or CLIP scores, TPIPS allows users to specify the specific dimension of comparison (e.g., "lighting", "pose of the person", "number of ducks") via free-form text, enabling more controllable and fine-grained image evaluation. The code, models, and data have been open-sourced.
Related event: CMU and Adobe Introduce TPIPS, a Text-Prompted Image Similarity Metric(2 posts)→
More from Multimodal
- ComfyUI-PETSCI I: Open-source node converts video to C64 style — PurzBeats · 2026-08-20
- Minimax H3 Local Generation: Character Dialogue and Lip Sync Test — TigerClaw305 · 2026-08-20
- Minimax H3 Test: Ref2v Workflow for Warhammer Scenes — tammy_orbit · 2026-08-20
- Mirelo releases major updates to Audio-to-MIDI tool — aziz4ai · 2026-08-20
- Meta's Muse Video Model Tested: High-Quality 10s Generation with Native Audio — testingcatalog · 2026-08-20
- Open-source AI short film 'ONEIRIC' releases all prompts and assets — ZeroStateReflex · 2026-08-20