Company
Rapidata
Swiss mobile-app crowdsourced human feedback; published video preference datasets such as Veo 3.1
Overview
- HQ: Zurich, Switzerland
- Funding: $8.5M seed (2026-02, co-led by Canaan Partners and IA Ventures)
- Links: https://rapidata.ai
This profile was compiled from the AI training-data column cards (Video generation); each column's data view is shown below under "Data profile" (in Chinese).
Data profile · 视频生成
Rapidata
瑞士移动 App 众包人类反馈,发布 Veo 3.1 等视频偏好集
- 数据类型
- 视频偏好标注;图像偏好标注
- 规模口径
- 触达 1500-2000 万人;约 150 万次标注/小时(公司口径,VentureBeat 转述,未复现)
做什么
通过第三方移动应用内的微任务收集人类反馈,把标注周期从数月压到数天。与 surge-ai、scale-ai 等专家型标注商处于两极。
规模
触达 1500-2000 万人,约 150 万次标注/小时,约 5500 人/分钟并发反馈(公司口径,VentureBeat 转述)[1]。种子轮 $8.5M(2026-02)[2]。研究基金最高给 $5 万标注额度/项目 [4]。
视频偏好集
公开发布的 Veo 3.1 偏好集在 HF 卡上标约 7.4 万条人类反馈、约 2.3 万标注者,题目为"Which video do you prefer aesthetically?",覆盖 Veo 3.1、Veo 2、Marey、Kling v2.1 等 [3]。FIRM-Video 用它构建约 3 万训练样本(仅据检索摘要)。
局限
美学偏好题较浅,多维度与物理一致性判断仍需专家。视频业务占比与客户名单大多未披露。
- https://venturebeat.com/data/rapidata-emerges-to-shorten-ai-model-development-cycles-from-months-to-days (2026-10-15)
- https://pulse2.com/rapidata-8-5-million-closed-for-scaling-global-human-feedback-network/ (2026-10-15)
- https://ai.gitcode.com/hf_mirrors/Rapidata/text-2-video-human-preferences-veo3.1/blob/main/README.md (2026-10-15)
- https://rapidata.ai/research-grants (2026-10-15)
Last compiled: 2026-10-15