Research and data
Tovaki Research publishes versioned data about the facts measured while integrating AI video models and checking delivery requirements for major publishing platforms. Every file states its scope, review date, and evidence boundary.
Catalog reviewed 2026-08-25. Each dataset keeps its own source review date.
This catalog contains two public datasets and one prepared benchmark method: a snapshot of Tovaki credit charges and integration-tested limits, a row-level record of delivery requirements from six platforms, and a pre-result contract for a controlled AI video comparison. JSON supports software and research workflows; CSV supports spreadsheets and analysis. No quality ranking has been published.
The files below are free to access. Counts and dates are generated from the same source modules as the human-readable pages so a page and its download cannot silently drift apart.
20 model rows · reviewed 2026-08-25
A dated snapshot of real Tovaki credit charges, supported durations, available resolution tiers, cost per finished second, and whether the Tovaki integration records a wired native-audio path.
6 platforms · 209 rows · 196 verified · reviewed 2026-07-25
Row-level publishing requirements for YouTube, Shopify, Amazon, TikTok, Instagram, and Xiaohongshu. Each row keeps its first-party source, document language, verification status, and review date.
A method record states what will remain fixed, what stays blind, how failures count, and when a result is allowed. Publishing it first prevents the outcome from rewriting the test.
Prepared · not executed · no ranking
Four integrations, six task lanes, 24 first-take cells and three blinded reviewers. The public record includes score weights, native-audio fairness, technical retry limits, score locking, publication boundaries and a SHA-256 commitment to the complete private protocol.
Model fields describe what Tovaki has integrated and exposed. A blank or false field can mean the capability is not wired in Tovaki, not that the vendor cannot provide it elsewhere.
The catalog does not rank visual quality because Tovaki has not published a controlled, repeatable quality benchmark for these models. Cost and technical limits are not a proxy for artistic quality.
These files contain no user count, market-share, award, rating, or best-in-class claim. Those statements require separate evidence and do not belong in an integration dataset.
Use the page that owns the fact. The current source wins when a dated snapshot and a live value differ.
Name Tovaki as the publisher, identify the dataset, include its review date, and link the exact JSON or CSV URL. When quoting a platform requirement, also retain the row's first-party source URL and verification status. Do not remove the integration or date qualifier.