RecipeBook by Shofo

Tentatively worth considering, but the evidence is thin. It is better treated as an initial sourcing option for video training data, not as a broadly validated data infrastructure product. The only available evidence is a Product Hunt listing saying it offers video training data by the hour with 25M+ clips. That supports the claimed positioning, but not the quality of the data, labeling depth, licensing coverage, or delivery reliability.

Based on the current evidence, its practical role is not video generation, video editing, or model hosting. A more accurate comparison is a dataset marketplace or data procurement service rather than an open repository like Hugging Face. If you build models for video understanding, retrieval, detection, or action recognition, it could in theory reduce sourcing time. But whether the data is directly training-ready still depends on metadata quality, cleaning, deduplication, licensing, and delivery format, none of which are confirmed by the evidence provided.

On cost and adoption barriers, the only supportable pricing signal is the official wording “buy by the hour.