xAI announced the launch of Grok Imagine Video 1.5 agent, integrating programmable agent workflows with image and video models to plan narratives and stitch together multi-camera shots. The driving layer has been upgraded to the latest Image 2.0, with official claims of higher image quality, more complete narratives, and stronger continuity between characters and scenes across multiple shots.

Video 1.5 is not an entirely new foundation. The model was fully opened in June via the Imagine API, with an accelerated version called Fast: generating 6 seconds of 720p video in about 25 seconds, compared to over 40 seconds for similar models. By the end of July, it added pure text-to-video generation, image and sound references, and native 1080p. Up to seven elements can be locked in reference images, and sound references are used to fix the same face and voice. Sound effects, ambient sounds, and dialogues can be generated in a single inference with the visuals.

The agent layer aims to address the issue of insufficient single 15-second segments: first breaking down story beats, then generating by shots, and finally stitching multiple segments into one. The official demonstration emphasized that "the continuity is better when connecting multiple shots." Product manager Andrew Milich described it as the planning capability of coding agents combined with the generative capabilities of image and video models, aimed at creating narrative with judgment rather than producing a random short clip all at once.

Image 2.0 handles 1K or 2K still frames, emphasizing layout, composition, and instruction following, with partition editing and segmentation tools. The video API is priced based on duration and resolution, with public rates approximately $0.08 per second for 480p, $0.14 per second for 720p, and $0.25 per second for 1080p, plus fees for reference image inputs. Watermarks on the consumer side cannot be removed. Access is available at grok.com/imagine as well as on iOS and Android.

In comparison with competitors, Video 1.5 had previously ranked high in the image-to-video category during its preview period, and subsequent public evaluations showed narrowing gaps with models like Gemini Omni Flash. OpenAI's Sora had once contracted due to resource and business model issues, while Google’s Veo, Seedance, and Tongyi Wanxiang continue to compete in the same short video space for creator workflows. xAI emphasizes the difference in being an "agent director" rather than just a single model scoring.

In terms of market mechanisms, buyers are creators with high demands for short dramas, advertisements, and character consistency, while sellers are generating APIs billed by the second. Funding has shifted from one-time prompt trials to reusable character references, voice locks, and project-level multi-agent parallelism. Beneficiaries are platforms that can sell 15-second slices as continuous narratives; those under pressure are old tools that only sell single shots without face and voice locking. The event-driven clarity is that the model is already available, but what is missing is the scheduling layer that can connect shots into stories.

On a supplementary note, the official has not disclosed the maximum number of shots an agent can plan at once, the longest duration of a completed film, or the interface boundaries between Image 2.0 and Video 1.5 in reference video editing; third-party tools have already begun to integrate Video 1.5 and Image 2.0.

Source: Public Information