VidScribe is a new for evaluating how well video generation models render and edit visual text. It covers four tasks: generating text from language prompts, transferring text from a reference image, sustaining text under motion, and editing text in place. The benchmark includes 803 human-verified samples designed to isolate 12 factors spanning text properties, imaging conditions, and temporal behavior, with 11 shared metrics and 2 task-specific probes.
Video generators preserve existing text better than they edit it
VidScribe tests text rendering across four video-generation tasks and finds that readable content, correct glyph shapes and reliable editing are distinct capabilities.
Alibaba Group · Shanghai Jiao Tong University · Fudan University
In one line
Video text generation capability is non-monolithic: I2V sustains text best, V2V editing is weakest, and degradation concentrates on structural and temporal factors.
Before you read this
Three ideas this paper builds onVidScribe covers four regimes: generating written text from a language prompt, transferring text from a visual reference, preserving text during image-to-video generation, and editing text within an existing video.
How it works
before and afterThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.