Grok Imagine 1.5 adds 7-image references, voice lock, and 1080p API
Text-to-video, character and voice consistency, and 1080p land in the xAI API — as California's AI-content law takes effect and OpenAI teases Astra.
Copy markdown
References, voice, and 1080p — now callable
Grok Imagine Video 1.5 now accepts up to seven image references to lock faces, products, and locations across scenes, plus a voice reference so a character keeps the same face and voice in every shot. Text-to-video and native 1080p are live through the xAI API on model grok-imagine-video-1.5.
Where to run it today
Text-to-video, image references, and 1080p are generally available on grok.com/imagine, iOS, and Android and callable from the xAI API; Replicate also hosts the model for image-to-video with synced audio. Image and voice references rolled out first to SuperGrok Heavy and Plus in the US, widening to all tiers within days.
The catch: voice API is gated
Voice reference is 'available on request' rather than open in the API, and the reference rollout started US-only on the top SuperGrok tiers. Replicate's card still lists the older 480p/720p image-to-video preview, so confirm you're on the 1080p path before you wire it in.
California's AI-content law goes live today
SB 942, the California AI Transparency Act, takes effect August 2: any tool with 1M+ monthly California users must embed latent metadata marking AI-generated images, video, and audio, ship a free public detection tool, and offer a visible-disclosure option. Penalties are $5,000 per violation, counted per day.
Elsewhere: OpenAI's Astra posts 10 Lean-checked proofs
OpenAI says an internal model, Astra, cracked ten previously open math and CS problems and formalized each in Lean, with certificates at github.com/openai/ten-proofs. Striking work — but there's nothing to call: Astra is unreleased, with no API and no launch date.