HiDream.ai Launches vivago R1 Globally, Upgrades Chinese Version Gouda

R1 first appeared at WAIC 2026 in July, where HiDream.ai presented it as the world's first unlimited-length content-creation agent. The system can generate a five-minute high-quality video in a single pass and supports unlimited multi-round extensions, the company said. That compares with the current industry mainstream of 15- to 30-second clips, which Leiphone noted began with Sora's 15-second demonstrations.

The difficulty with longer videos lies in keeping a story coherent across multiple shots, characters, and scenes, not simply extending duration. R1 uses an Agent system to plan and schedule long tasks: a main Agent breaks down the creative idea, while sub-Agents work on script, storyboard, characters, scenes, and sound, according to the report. The company says this keeps character appearance, scene style, and narrative rhythm consistent across segments. With what HiDream.ai calls HD-AgentOS, its scheduling and governance layer, R1 raises the effective usable success rate of content to 85%, reducing the need for repeated retries. The system also improves cross-shot character consistency and scene transitions, the company said, softening the "plastic" and "teleporting" effects common in AI video and making camera movement and audio-visual alignment more natural.

R1 adopts a Chat-First approach. HiDream.ai argues that video creation should start with what a creator wants to express rather than with nodes, canvases, and parameters. Current AI video tools largely follow two routes: canvas- or node-based workflows that give users high freedom but also high barriers, and conversational tools like R1, where users describe needs in natural language and adjust through dialogue. Behind the experience are two principles the company calls "doing the back" and "doing the thick": encoding professional creators' judgment, aesthetics, and experience into Agent orchestration, and building a SkillHub that translates abstract AI capabilities into real scenarios such as product ads, character videos, story shorts, and brand content. In the full workflow, users express a goal, Agents break it down and execute, and SkillHub matches and arranges the needed skills into an executable task sequence.

The foundation for this is HiDream.ai's native omni-modal world model, which the company says provides multimodal understanding and generation. It places text, images, video, reference assets, documents, and previous creation results into the same task context so Agents can understand the whole task rather than isolated instructions. The Agent system then organizes model capabilities into a continuing creation process, planning across script, storyboard, characters, scenes, shots, and sound. R1 also offers an asset library where users can store generated, selected, and approved content for reuse in later projects.

At an immersive workshop for vivago R1 and Gouda, creators from different fields shared their experiences online and offline. One creator used R1 to rework the classic Journey to the West into a nearly five-minute piece, demonstrating the system's visual translation of a classic text. Another creator presented an inspirational short about physical education teacher Zhou Lan, who moves from hesitation to starting again; the report said details such as muscle tremor while running, sweat highlights, and the texture of footsteps on a rubber track were rendered closely. In a commercial case, an AI creator showed a red foldable phone advertisement in which R1, based on a prompt, understood the relationships among product, people, and scenes and generated multiple phone views, a female designer's styling, studio and outdoor life scenes, and visual assets including bags, color cards, and sketches. An overseas creator said, according to the report, "The endpoint of video creation is an expression."

HiDream.ai also introduced CHATS, a framework for evaluating whether a video Agent can deliver finished works. Its five dimensions are Consistency, for stable characters, scenes, and styles across shots; Human Intent, for understanding creative ideas and advancing them over multiple dialogue turns; Audio & Visual completeness, for organizing sound, rhythm, and the overall audiovisual experience; Timeline & Tale, for sustaining a story across shots and sections; and Stability, for reducing repeated generation and rework while raising usable content rates. The company said these dimensions were repeatedly tested by internal users in real creation scenarios.

Vivago.ai was launched in May 2024 and became one of the world's earliest publicly available AI video generation products, according to the report. In May 2026, it topped Product Hunt's daily list, and its global users have now surpassed 70 million. HiDream.ai says the "R" in vivago R1 stands for Reasoning, using deep inference to turn inspiration into works, and Rock, a nod to "The Show Must Go On."

Editor's Summary

HiDream.ai launched vivago R1 globally on Sept. 8, 2026, and upgraded its Chinese version Gouda, positioning the AI video agent around five-minute single-pass generation and unlimited extension. The company also introduced the CHATS evaluation framework and cited more than 70 million vivago.ai users. The release reflects an industry push from short clip generation toward longer, more consistent and deliverable AI video projects.