APIMODELSAPIMODELS
APIドキュメント

apimodels.app

Grok Imagine Video 1.5 API

xAI's Grok Imagine video model: text-to-video and image-to-video under one model name, with a native synced audio track (dialogue / sound effects / ambience), any duration from 1 to 15 seconds, three per-second pricing tiers (480p / 720p / 1080p), and watermark-free output. Generation is fast — a 4-second clip comes back in roughly 30-70 seconds in our production tests. It runs on the same unified video endpoint as VEO, Kling and Seedance on apimodels.app, so switching models is a one-string change and one API key covers everything.

Pricing (per second)

ResolutionRate4s8s15s
480p$0.045/s$0.18$0.36$0.68
720p$0.0529/s$0.21$0.42$0.79
1080p$0.0882/s$0.35$0.71$1.32

Special price until 30 November 2026 (Beijing time): 40% off at $0.027 / $0.0317 / $0.0529 per second for 480p / 720p / 1080p, priced at the time the job is created.

Cost = rate × duration in seconds. Text-to-video and image-to-video cost the same; reference images add nothing. Failed tasks are not billed.

Create a task

POST /api/v1/video/generations — async: create, then poll the same endpoint for the result.

# Step 1: Create task (text-to-video; add "images" for image-to-video)
curl -X POST https://api.apimodels.app/v1/video/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-imagine-video-1.5",
    "prompt": "A calm coastal shoreline at sunset, camera slowly pushing forward",
    "resolution": "720p",
    "duration": 10,
    "aspect_ratio": "16:9"
  }'

# Step 2: Poll status
curl "https://api.apimodels.app/v1/video/generations?task_id=TASK_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

Request parameters

FieldRequiredTypeDescription
modelYesstringgrok-imagine-video-1.5
promptNostringText description (optional when reference images are given — send at least one of prompt / images). Up to 4096 characters; for 1080p clips of 10-15s keep it within 2048, longer prompts are served by the other channel and render slower
imagesNostring[]Reference images, up to 7; http(s) URLs or base64 / data URLs both work. At 1080p only one image is accepted (used as the start frame); for 2+ images use 720p or 480p
resolutionNostring"480p", "720p" (default), "1080p". 1080p is available for text-to-video and single-image image-to-video only; multi-image reference-to-video is capped at 720p by the model
durationNonumber1-15 seconds, default 10
aspect_ratioNostring"16:9" (default), "9:16", "1:1", "3:2", "2:3"
callback_urlNostringWebhook URL called when the task completes

Getting the result

The create call returns data.taskId right away and the video renders in the background. GET the same endpoint with that task_id and read data.state: pending means still running, completed puts the MP4 URL (H.264 video + AAC audio) in data.resultUrls[0], and failed puts the reason in data.failMsg. To skip polling, pass callback_url and we POST you when the task finishes.

  • Measured turnaround: a 4-second clip returns in roughly 30-70 seconds (text-to-video is faster, image-to-video slightly slower).
  • Result videos are kept for 30 days and then deleted — download or re-host anything you want to keep.
  • You are charged on success; failed tasks are not billed.

Things worth knowing

  • Native audio: the model produces a synced audio track (dialogue / sound effects / ambience) alongside the frames — no separate dubbing step. Measured 480p output is 848×480 at 24fps.
  • Output is watermark-free — we frame-checked both text-to-video and image-to-video results.
  • Image-to-video is faithful to the reference: composition, subject and materials are preserved, and the prompt describes the motion and camera you want.
  • The 1080p rule with images comes from the model itself: multi-image reference-to-video is capped at 720p, and only single-image start-frame image-to-video renders at 1080p. So at 1080p send exactly one image and the clip starts from it; omit aspect_ratio to follow the image (a portrait image yields a portrait clip), or set it explicitly. Two or more images at 1080p return a 400 at creation — switch to 720p.
  • At 480p / 720p images are handled in reference mode (subject and composition preserved, up to 7 images) — unchanged.
  • Reference images can be sent as raw base64 — we automatically stage them to an accessible URL before generation, identical to sending a URL.
  • Reference-video editing (video-to-video) is not enabled yet; sending a reference video returns an explicit 400 rather than being silently ignored.
  • Auth is the same as every other model: send Authorization: Bearer YOUR_API_KEY. One key covers every video, image, audio and LLM model on the platform.

Limits and gotchas

Every row below is a hard limit of the model or its upstream, not something the platform adds; requests that exceed one are rejected with a 400 at creation (no task, no charge) instead of failing mid-way.

ItemLimitNotes
Prompt lengthUp to 4096 charactersCounted in characters (CJK and Latin count the same). 1080p requests of 10-15 seconds are served first by the per-clip channel, which accepts at most 2048 characters; longer prompts still render but move to the per-second channel and take longer. Over 4096 is a 400.
Prompt contentOnly what the model needsStoryboard beats, camera, dialogue, style and character looks are effective; instructions written for humans ("do not read the notes below", "output spec", "language policy") are not understood by the model — they only burn the character budget and dilute the real directions. To get a specific length set the duration parameter; do not write "exactly 15 seconds" in prose.
Reference imagesUp to 7; exactly 1 at 1080p480p / 720p use reference mode (subject and composition preserved). 1080p only has single-image start-frame mode: the clip starts from that image; two or more images return a 400 — switch to 720p.
Reference image reachabilityMust be downloadable at submissionWe copy the image to our own storage the moment the task is created, so your link expiring afterwards does not affect generation. Short-lived signed URLs (a few minutes) are fine as long as they are still valid when you call; if unsure, send base64.
DurationInteger 1-15 s, default 10Output length equals the requested duration. 1080p at 10-15 seconds is the most affordable combination (per-clip pricing, below the per-second rate); 1080p under 10 seconds is billed per second.
Resolution480p / 720p / 1080pOnly these three lowercase values. See the 1080p image rule above; the 720p cap for multi-image reference is a model limit no channel can bypass.
Aspect ratio16:9 / 9:16 / 1:1 / 3:2 / 2:3In 1080p single-image mode, omitting aspect_ratio follows the reference image; otherwise the default is 16:9.
Reference videoNot enabledSending video / video_url etc. returns an explicit 400.
AudioGenerated natively with the framesPut the dialogue language and lines in the prompt; uploading audio or choosing a voice is not supported, and per-line lip-sync timing is not guaranteed.
Turnaround~30-70 s for a 4-second clip; ~2.5-3 min for 1080p 10-15 sMeasured medians; longer at peak times. Results are kept for 30 days.

Common 400 errors

Parameter problems come back synchronously as a 400 at creation ({"code": 400, "msg": "..."}) — no task is created and nothing is charged. The msg strings are verbatim below, safe to match on.

msgCause / fix
prompt or a reference image is requiredNeither prompt nor images was sent; provide at least one.
prompt is N characters; at most 4096 are accepted on this model — shorten itPrompt longer than 4096 characters. Storyboards, dialogue and production notes add up fast — trimming to 2048 or fewer is the safe target.
resolution must be one of 480p / 720p / 1080p (got "...")Unrecognized resolution; only these three lowercase values are accepted.
duration must be 1-15 seconds (got ...)Duration outside 1-15 seconds.
at most 7 reference images are supported (got N)More than 7 reference images.
1080p supports text-to-video and single-image image-to-video only; multi-image reference-to-video is capped at 720p by the model (got N images) — set "resolution": "720p" or send a single imageTwo or more images at 1080p. Set resolution to 720p, or keep a single image.
reference-video editing is not enabled on this model yet — pass reference images insteadA reference video was sent (video / video_url etc.); video editing is not enabled on this model yet.

Tasks that fail during generation (upstream error, timeout, etc.) end in state failed with the reason in data.failMsg, and the charge is refunded automatically.

Related

FAQ

Is there an API for Grok Imagine video generation?

Yes. On apimodels.app you POST to /api/v1/video/generations with model "grok-imagine-video-1.5", a prompt and/or reference images, then poll the same endpoint with the returned task_id until data.state is "completed" — data.resultUrls[0] is the MP4 (H.264 video + AAC audio). It is the same request shape as VEO, Kling and Seedance here, so switching models is a one-string change, and one API key covers all of them.

How much does Grok Imagine Video 1.5 cost?

It is billed per second by resolution: 480p at $0.045/s, 720p at $0.0529/s, 1080p at $0.0882/s — so a 4-second 480p clip is about $0.18 and a 15-second 1080p clip about $1.32. Note the defaults: omitting resolution and duration gives you 720p and 10 seconds (about $0.53), because that is the shape both of our channels can serve — ask for 480p or a shorter clip explicitly and you get it, it just runs on a single channel. Text-to-video and image-to-video cost the same; reference images add nothing. Failed tasks are not billed.

Does Grok Imagine Video generate audio, and is there a watermark?

It produces a native synced audio track (dialogue, sound effects, ambience) alongside the frames — the MP4 you download already carries H.264 video plus AAC audio, no separate dubbing step. Output is watermark-free; we frame-checked both text-to-video and image-to-video results in production before listing it.