Grok Imagine video in Sume: why it needs a start-frame image

Grok Imagine in Sume's Agents Videos panel needs a start-frame image: submit without one and the panel stops with the message "Grok Imagine requires a start-frame image." It is image-to-video only, offers 5, 6, 8 or 10 seconds at 480p or 720p, and sends neither audio nor aspect ratio.

The limit is in Sume's panel, not in xAI's model. xAI's video generation guide lists grok-imagine-video-1.5, describes both text-to-video and animating a still image, and says duration can run up to 15 seconds. Sume's panel wires only the image-to-video path and stops at 10 seconds.

What does the panel send for Grok Imagine?

When you pick Grok Imagine, the panel maps it to the grok-imagine-video-1.5 model id and builds a smaller request than for other models. Aspect ratio is left out, audio is left out, and no end image is sent. Only the prompt, duration, resolution and the start-frame URL travel.

That matters if you are used to setting a 9:16 aspect ratio on other models. For Grok, prepare the start frame in the shape you want, because the panel will not ask the model for a ratio.

Can I do text-to-video with Grok Imagine on Sume?

Not through this panel, even though xAI describes text-to-video. Without a start frame the submit is blocked before it reaches the API. If you need text-only prompts, pick Auto, Wan 3.0, MiniMax H3 or Kling 3.0, which accept an optional start frame rather than requiring one.

The existing post on Grok Imagine 1.5 text-to-video versus Sume's image-to-video-only row goes further into the vendor side.

When is Grok Imagine the right pick?

Pick it when you already have a strong still, such as a product shot or a character frame, and want expressive motion from it for a short clip. For a hero image that must be preserved, an image-to-video model with a start frame keeps the first frame anchored.

Pick something else when you need audio, a fixed last frame, references, or more than 10 seconds. Sume's docs say every catalog model other than the long-clip rows tops out at 15 seconds, and several of the longer ones are covered in which AI video model for a 15-second single take.

• Start frame in hand, short clip, no audio needed: Grok Imagine.
• Need audio: Wan 3.0, MiniMax H3 or H3 Max.
• Need a last frame: Wan 3.0 or MiniMax H3.
• Need to run from text only: not Grok Imagine.

How does it compare with the other panel models?

Within the same panel, Kling 3.0 offers 5 to 15 seconds up to 4K, Wan 3.0 runs 2 to 30 seconds, and MiniMax H3 runs 5 to 15 seconds at native 768p with stereo audio. Grok Imagine has the lowest top resolution and a 10-second ceiling in that set, which is why it suits quick motion tests from a still rather than final delivery.

A practical pattern is to test a still with Grok Imagine at 480p, decide whether the motion idea works, then move to a model with a last frame or audio for the keeper. See Seedance versus Grok Imagine for the other direction of the comparison.

Sume does not publish a benchmark ranking between these models and this post does not either; the table above is a control comparison, not a quality one.

What should I check before submitting?

Check that the start-frame URL is reachable. The API takes public HTTPS media URLs, and the docs list inaccessible reference images as a first thing to check when a generation fails. See the video generation docs for the troubleshooting list.

Billing follows the rest of the video catalog: Sume reserves provider list times 1.25 on submit, and the poll response's usage.cost is the billable amount. Read it from the job instead of estimating from the panel.