Grok Imagine <IMAGE_0> tags shift with a first frame; Sume is 0-based

On xAI, if you also set image to pin the first frame, that image takes <IMAGE_0> and your reference_images start at <IMAGE_1>. On Sume's Omni route (gemini-omni-flash-1.1), tags are <IMAGE_REF_0> and <VIDEO_REF_0>, 0-based in the order of the reference lists, and the docs do not describe a first-frame shift.

Sources: xAI's reference-to-video page and the Video Router docs, read 2026-10-01.

How does xAI number reference images?

The page says to refer to reference images by their position in the list, starting at <IMAGE_0>, then <IMAGE_1>, and so on. With image set as well, the first frame takes <IMAGE_0> and the references start at <IMAGE_1>. Voices are tagged <AUDIO_0>, <AUDIO_1> and <AUDIO_2>.

How does Sume number references?

The Omni row says to address media as <IMAGE_REF_0> and <VIDEO_REF_0>, 0-based, in list order. Caps are reference_image_urls up to 10 and reference_video_urls up to 3, each 3 seconds or shorter.

Can I combine a video_url with references on Sume?

No. The docs say video_url is the edit source, not a reference, and it cannot be combined with image_url, end_image_url or reference_*_urls. An edit request therefore has no tags to number.

How do I avoid an off-by-one tag?

Number from the first list you send, and test one request with a single reference before a large one. If you port a prompt written for xAI's first-frame case, subtract the shift yourself: every <IMAGE_n> that followed the pinned frame becomes <IMAGE_REF_n-1>, but only when your Sume request sends references alone. For the same tag idea on another model, see Seedance 2.5 Omni reference at-tags.