Realtime AI News
Genspark launches Media Studio, a video agent that orchestrates 70+ models
Genspark has launched Genspark Media Studio, an all-in-one video agent workbench that pairs director models like DeepSeek and Claude with more than 70 generation models from over 20 vendors. A Story Bible locks character designs, clips can be selectively reshot, and members get 50% off credits until November 4.
Genspark has launched Genspark Media Studio, a video agent workbench it bills as all-in-one: scripting, storyboarding, character design, asset generation and editing all happen inside the same canvas and conversation, so users with no editing background can walk away with a finished video.
Per the company's own description and QbitAI's hands-on test, the system uses a large language model as the "director brain" that interprets the request, writes the script and assigns work to different models. Users pick from three director options — DeepSeek V4.1 Flash, Claude Opus 5.5 and Claude Fable 5.1 — while Seedance 2.5, Kling and Veo handle video, GPT Image and Nano Banana handle images, and ElevenLabs and others handle audio. In total the platform connects more than 70 model versions from over 20 vendors.
The differentiator is orchestration rather than any single model. Genspark packages production flows as skills — one for ads, another for narrative shorts — and the agent picks the right one for the task. For a drama it writes the script, breaks it into shots, locks characters and locations, and only then hands the footage to video models.
A Story Bible stores those decisions: character appearance, wardrobe and props are fixed as reusable assets before shooting starts, and the same cast and world can be carried into other projects. In the test, an initial five scenes were split into 84 shots grouped into 12 video units, after which the agent checked results against the script and settings and scheduled reshoots on its own.
Editing is addressable rather than all-or-nothing. Reshoot clips lets a user mark an unsatisfactory stretch on the timeline and regenerate it, up to five segments per pass at four seconds or more each, while Frame edit can box an object and replace or remove it in videos under 30 seconds. Drafts render at 480p and are upgraded to 1080p only after approval, which keeps fixes in the cheap stage.
Knowledge explainers and drama shorts take different technical routes. Knowledge Explainer leans on code-to-video: the model writes code that lays out text, charts and animation and then renders it, and Genspark says the rendering itself consumes no credits, with narration and a few atmospheric shots billed separately. Narrative work instead offers a 3D white-model preview that locks blocking and camera angles before Seedance generates footage.
Pricing is part of the pitch. Plus and Pro members get 50% off credit consumption inside Media Studio from 18:00 Beijing time on October 27 to 18:00 on November 4, covering planning, image, video, audio and 3D generation. The tester burned roughly 45,000 of an initial 50,000 credits across three projects — a short drama, an explainer and a pixel-art mini game — a reminder that cost still bounds daily use.
Coordinating more than 70 model versions shifts the platform's selling point from raw generation quality to scheduling reliability. Multi-model orchestration is becoming the competitive edge in video agents, and the questions that matter next are whether sync drift and dropped frames can be tamed, and whether credit costs fall far enough for ordinary creators to use this every day.
Why it matters
The launch moves video generation competition from raw model quality to orchestration and production engineering, lowering the barrier for shorts, ads and explainers; for model vendors, a place inside such workbenches now translates directly into usage.
Nearby Updates
All10/29, 15:00
Ropes & Gray Builds a Transactional Diligence Tool With OpenAI
Ropes & Gray has built a proprietary AI diligence tool with OpenAI that reads thousands of pages, groups contracts by theme and issue, and drafts a report in the firm's own format. Lawyers still review every summary and source document, and the firm aims to cut issue-level diligence from weeks to hours.
10/29, 15:45
AI agent startup TinyFish opens Vietnam office to house core agent work
Silicon Valley-based TinyFish has opened an office in Vietnam to carry out core AI agent development work, according to TNGlobal. The report does not disclose the city, team size or investment behind the new site.
10/29, 16:59
Unsloth ships Browser use, letting Qwen3.8-27B drive a browser locally
Unsloth has released a tool called Browser use that lets Qwen3.8-27B control a browser locally, according to a report carried via Google News. Keeping the agent on-device takes sessions and credentials off external APIs, though reliability on real pages remains unproven.
10/29, 17:00
OpenAI explains how layered monitoring catches frontier model misalignment
OpenAI has detailed the layered monitoring it uses to catch frontier model misalignment and unauthorized agent behavior before incidents escalate. The post describes how responders are meant to intervene early rather than after damage is done.