Runway's 24 August changelog says Wan 3.0 is available in Tool Mode and Workflows, accepts multiple image, video and audio references, generates video with audio, and offers 480p, 720p and 1080p. Days earlier, Runway added MCP support for listing, opening, editing and running workspace workflows. Together, those changes make a model more than a manual selection; it can become a repeatable production node called by an agent.
Wan 3.0 is now inside Tool Mode and Workflows
Runway's 24 August changelog says Wan 3.0 is available in Tool Mode and Workflows, accepts multiple image, video and audio references, generates video with audio, and offers 480p, 720p and 1080p. Days earlier, Runway added MCP support for listing, opening, editing and running workspace workflows. Together, those changes make a model more than a manual selection; it can become a repeatable production node called by an agent.
Every call needs a shot contract
A shot contract should lock five categories: call date and product surface, visible model name and version, hashes and permissions for every reference, resolution and audio settings, and the reason for acceptance or rejection. If an agent can edit the workflow, node changes and the actor who executed them also belong in the record. When the same prompt later produces a different result, the team can separate platform change from operator error and decide whether to keep the old shot, rerun or change models.
An available surface does not prove accepted-shot cost
Before production delivery, the workflow needs a title-specific handoff. Keep benchmark inputs for character close-ups, action, environments and dialogue, then record generation count, queue time, human repair, rejection reason and final version. After a node changes, rerun this small suite before reopening season-scale jobs. That separates product availability from project readiness and leaves a practical fallback when a platform rolls back, removes a model or changes pricing.
What remains unknown: Before production delivery, the workflow needs a title-specific handoff. Keep benchmark inputs for character close-ups, action, environments and dialogue, then record generation count, queue time, human repair, rejection reason and final version. After a node changes, rerun this small suite before reopening season-scale jobs. That separates product availability from project readiness and leaves a practical fallback when a platform rolls back, removes a model or changes pricing.