AI video tools are becoming studios, not just generators

From a clip to a production

Genspark says it has launched Media Studio, an all-in-one AI studio for video production that uses an agent-based director system to plan and execute video production. The system orchestrates generation models across image, video, audio and 3D categories, the company says, rather than asking a user to prompt a single model for a single clip. m:…2815

Media Studio automates scriptwriting, casting, sets, storyboarding, voice, score and editing, Genspark says, and Media Canvas provides a single interface for managing scripts, characters, sets and storyboards. The agent agrees on concept, script, storyboard and budget before rendering, the company says, and users can provide feedback by circling elements on a frame to request changes. m:…2815

Genspark says Claude Opus 5.5 and Fable 5.1 act as the director brain for planning and scripting, GPT-6 Astra handles multimodal verification in the 3D director stage, and the model roster includes Seedance 2.5, GPT Image, Nano Banana, ElevenLabs and Seed Audio. The Pro-level Editor includes a Photoshop-class layer editor, a multi-track video timeline, an AE-style Studio, a 3D white-model director stage and an interactive film and game engine, the company says, and the studio supports projects ranging from 15-second ads to 10-minute short dramas. m:…2815

Small tools inside a larger agent

Luma says it has launched eight specialized mini-apps within Luma Agents to perform specific creative tasks without requiring full image regeneration. It lists Luma Plus tier pricing starting at 30 dollars per month and Pro tier pricing at 90 dollars per month. m:…6373

Luma says Skills allow users to save multi-step workflows as reusable one-click tools, that Uni-1 powers its image intelligence, and that Product Shots generates three images per run in a consistent editorial style. Ray 3.2 provides production-ready video output with frame-by-frame creative control, the company says, and Luma provides an API for custom integrations with production pipelines. This suggests the studio model is not only about generating more, but about reusing approved steps without redoing approved work. m:…6373

Creative expertise, packaged as software

Runway says it has launched App Studio, a new way to build custom creative software inside Runway, allowing users to build custom creative software with no coding required. App Studio builds custom apps that generate images, video and audio using Brand Kit and Workflows, the company says, and it turns institutional knowledge into tools the whole company can run. m:…4470

App Studio is available to all paid plans, Runway says, and usage draws from existing Runway credit balances. Enterprise admins can disable specific models in apps and can see Apps Studio usage across every app in the workspace, the company says, and App Studio writes, tests and deploys every app, with code remaining visible for anyone who wants to inspect it. m:…4470

Runway says KPF used App Studio to build an app that brings design renderings to life, that public links for apps require sign-in with a Runway account, and that visitors using public links pay for their own generations. App Studio is launching with a series of built-in apps, the company says, and users can update workflows, references or Brand Kit at any time. The shift here is from sharing a prompt to sharing a tool that already contains the brand, the workflow and the controls. m:…4470

Model choice still matters inside the studio

A comparison published on fal shows why model choice remains a budget decision even inside a studio workflow. Wan 3.0 and Seedance 2.5 are both available for commercial use on fal, the comparison says, with Wan 3.0 billed per second of output at 0.05 to 0.20 dollars depending on resolution, and Seedance 2.5 billed per token at an estimated roughly 0.2205 to 1.164 dollars per second. m:…0303

Seedance 2.5 reference-to-video accepts up to 30 images and 50 files per request, the comparison says, editing and extension tasks for existing footage are specific to Seedance 2.5, and Seedance 2.5 supports lip-synced speech among its audio outputs. Input video duration is charged on both models, the comparison says, with Wan 3.0 adding input seconds to output seconds for billing, and Seedance 2.5 including input seconds in its token count before a 0.6 multiplier. m:…0303

The comparison finding, as summarized in the source material, is that Wan 3.0 favors cost-efficient volume while Seedance 2.5 excels in complex editing and reference-heavy tasks. For creators, the implication is to choose the studio or app first for workflow and brand control, then choose the model inside it for the job at hand, since a cheaper model for volume and a stronger model for editing solve different problems. m:…0303