Grok Imagine Video 1.5 added prompt-only and reference-guided video in July
The June 16 general-availability release exposed grok-imagine-video-1.5 as an image-to-video model in the xAI API. It took a starting image and a motion prompt. Video 1.5 Fast reached Grok’s web and mobile products separately, and xAI attributed its June speed claim of about 25 seconds for a six-second 720p clip to that variant. The earlier June 3 API preview used a preview model identifier, and May’s Quality Mode was an image-generation and editing release. June GA announcement · June preview · May image Quality Mode
On July 31, xAI announced that the same 1.5 API model could start from a text prompt or use images as references. With prompt-only generation, the service creates a still internally and animates it. The separate reference route can guide what a subject, product or setting looks like across a shot. The launch described up to seven references and native 1080p for text-to-video and image-to-video. Its API reference example used 720p, and the reference-to-video guide lists a 720p ceiling. July 31 announcement · API release notes · Reference-to-video guide
In the consumer products, xAI announced prompt-only video and native 1080p on web, iOS and Android. Image and voice references started on web and iOS for U.S. SuperGrok Heavy and Plus subscribers, with a wider rollout announced. The July product announcement offered API voice references on request. The API release notes now include optional preset voices under July 31. These records establish the announced route and leave the date of broad API preset access unresolved. July 31 announcement · API release notes
What operators observed
- August 6, Atlas Cloud. Fun_Walk_4965 used three character references, three environment references and a separate opening illustration for a 15-second anime riverbank scene. A generic request for dynamic movement tended to produce one slow camera move. Specifying changes in camera position, framing and shot size gave the sequence cuts, with some imperfect transitions. The route was Atlas Cloud. The post supplies no resolution, endpoint or voice settings. Original scene and workflow
- August 30, direct API through MCP. Junji Takashima’s repository documents a nominal 1080p text-to-video request that returned 1920×1088, plus separate reference-to-video validation with a preset speaker voice. The author puts reference-to-video at a 720p ceiling. The repository gives a test date and model but supplies no output file or request ID for those two claims. A dated directory listing confirms an August 30 update, while an immutable commit for the specific README wording was unavailable. Maintainer test notes · Voice prompt documentation · Dated directory listing
- November 2, mediated API. BDevs tested a fictional CLI-documentation pipeline. The named 1.5 model’s text-only clips spelled a command correctly in sampled frames at 480p and 720p, but the prompt’s typing action never appeared. The reference test preserved the source text in sampled frames until the requested zoom took part of the command out of view. Its file measured 736×400. The report provides a workflow observation from the mediated route without a native 1080p or preset voice test. BDevs tests 3 and 4
Pubroot’s earlier direct-API benchmark provides a dated baseline. The team published July 24 from a July 4 run and reported a five-second 1.5 image-to-video request at 1080p completing in 70.4 seconds wall time. Text-to-video was tested on the older model ID. The 1.5 Fast slug returned a 404 for that team. The benchmark’s private harness and ephemeral result URLs limit independent inspection. Pubroot benchmark
The July 24–August 14 Reddit window also includes integration notices from Venice and CharGen. They date those intermediary listings. Within this search window, the Atlas post supplies creator evidence for visual references, while distinct direct-API evidence for 1080p prompt-only output and preset-voice acceptance comes from the later August repository report. Venice notice · CharGen notice · Atlas creator workflow