DEV Community

AI Studio
AI Studio

Posted on

Don't Budget Grok Imagine 1.5 Off the API Card Alone

You are sizing a video generation ticket from the official Grok Imagine 1.5 and 1.5 Lite capability card: reference images, native 1080p, voice controls, pinned frames, and per-second API pricing. Before writing grok-imagine-video-1.5 into a pipeline spec or budget, split what the SpaceXSI (formerly xAI) API supports from what the surface you will actually call exposes.

Pick the Model by Constraint

The official SpaceXSI developer documentation (last updated 13 November 2026 and rechecked 18 November 2026) recommends a clear split across the two current models. Choose grok-imagine-video-1.5-lite for straightforward clips at the lowest per-second rate, and choose grok-imagine-video-1.5 when the job requires reference images, voice references, pinned frames, or native 1080p output.

Both models share a wide baseline in the Grok Imagine API. Each generates clips from 1 to 15 seconds at 480p, 720p, or 1080p with native audio and lip-synced speech. Both support text-to-video, where the model renders a first frame from the prompt and animates it, and image-to-video, where the uploaded still becomes the first frame and the output inherits that image's aspect ratio. For text-to-video, both accept nine aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9, and 5:2.

The differences appear as soon as a brief introduces delivery or continuity constraints:

  • Native 1080p vs upscaled 1080p. Although 1.5 Lite lists 1080p in its resolution options, its text-to-video and image-to-video 1080p outputs are upscaled from 720p. On 1.5, text-to-video and image-to-video render 1080p natively. If a client or downstream compositor enforces true 1080p pixels, 1.5 Lite misses that requirement despite listing the resolution.
  • Reference images. Only 1.5 supports multi-reference generation, accepting up to 14 reference images for clips up to 15 seconds at 720p. Reference inputs carry characters, products, and locations across a shot without locking the first frame, whereas image-to-video fixes frame zero and expects prompts to describe subsequent motion. Lite has no reference-image support.
  • Voice references and pinned frames. Up to 3 voice references (preset voices by default, with own-audio voice references restricted to trusted partners on request), first-and-last-frame pinning (including reusing one still at both ends to create a loop), and up to 4 mid-video keyframes exist exclusively on 1.5. Lite omits all four controls.
  • Video editing and extension. Neither 1.5 nor 1.5 Lite supports video editing or video extension in the API. Those operations remain on the older grok-imagine-video model. Any pipeline that generates a clip on 1.5 and extends it afterward must route the second call to grok-imagine-video.

Official SpaceXSI API pricing shows why defaulting blindly to 1.5 inflates estimates. Output on grok-imagine-video-1.5 costs $0.08 per second at 480p, $0.14 at 720p, and $0.25 at 1080p, plus roughly $0.01 per image input. By comparison, grok-imagine-video-1.5-lite starts at $0.02 per second at 480p, charges $0.03 at 720p, and reaches $0.14 at 1080p. At 720p, 1.5 costs nearly five times as much per second as Lite before counting input images.

One version check also matters for roadmaps. As of 18 November 2026, no Grok Imagine Video 2.0 has been released or announced; the model named "Grok Imagine Image 2.0" is a separate still-image model. Production designs built today target the 1.5 pair and grok-imagine-video.

Hosted UIs Expose a Subset of the API

A second planning error happens when teams read the SpaceXSI API card, build a budget around those capabilities, and then run part of the workflow inside a hosted browser interface rather than calling the API directly. Hosted multi-model frontends frequently expose a curated subset of an underlying model's parameters.

Take AI Studio as a concrete example. It is an independent website that runs several models in the browser (including Seedance, Kling, Gemini Omni, Happy Horse, MiniMax H3, and Grok Imagine), and it has no affiliation with SpaceXSI or xAI. Its production interface was checked on 11 November 2026, and the SpaceXSI API capability card rechecked on 18 November 2026 showed no changes.

On AI Studio's video generator page, both models sit under the "Grok Imagine Video" family switcher, with "Grok Imagine 1.5 Lite" listed first as the family default alongside "Grok Imagine 1.5". Every clip generated there runs from 4 to 15 seconds in 1-second steps (defaulting to 5 seconds), offers 480p, 720p, or 1080p (defaulting to 720p), and ships with native audio with no mute toggle.

Compared directly against the official Grok Imagine API card, the hosted UI narrows what each model accepts:

  • On AI Studio, 1.5 Lite supports text-to-video (in 16:9, 9:16, 1:1, 4:3, 3:4, 3:2, and 2:3, with 16:9 default) and single-image first-frame image-to-video.
  • On AI Studio, 1.5 supports text-to-video (16:9, 9:16, 1:1, 3:2, and 2:3), single-image first-frame image-to-video, and reference-to-video using 2 to 7 reference images capped at 720p (480p or 720p only).
  • Several API and first-party app capabilities do not appear on AI Studio at all: voice references and preset voice selection, first-and-last-frame pinning, mid-video keyframes, 8 to 14 reference images, 21:9 and 5:2 aspect ratios, video editing and extension, Projects or agent tooling, and durations below 4 seconds.

Billing units also diverge between direct API calls and a hosted site. AI Studio charges credits per second of output, computed as rate times seconds and rounded up, with no extra credit charge for reference images. The site displays the exact credit total before you generate. On 1.5 Lite, rates are 1.5 credits/s at 480p, 2.5 at 720p, and 10 at 1080p. On 1.5, rates jump to 6 credits/s at 480p, 10 at 720p, and 18 at 1080p.

One quick comparison shows the cost swing in practice. A 5-second clip at 720p costs 13 credits on 1.5 Lite (12.5 rounded up) versus 50 credits on 1.5. Selecting 1.5 on the hosted UI nearly quadruples the credit burn per clip at 720p, while still leaving out API-only features such as voice references, keyframes, and 14-image packs. Those capabilities require SpaceXSI's own app or API.

Pre-Commit Checklist for Pipelines and Estimates

Run these checks before locking a model string, clip length, or unit price into an engineering ticket:

  1. Verify the live SpaceXSI API docs and pricing page before finalizing specs, and plan around the 1.5 family plus grok-imagine-video since no Grok Imagine Video 2.0 exists as of 18 November 2026.
  2. Justify grok-imagine-video-1.5 with a specific constraint: native 1080p, reference images, voice references, first-and-last-frame pinning, or mid-video keyframes. Route simple text-to-video or first-frame image-to-video jobs to grok-imagine-video-1.5-lite.
  3. Inspect the actual calling surface before promising counts or controls. Do not schedule voice references, keyframes, sub-4-second clips, 21:9 or 5:2 ratios, or 8-to-14-image reference packs on AI Studio.
  4. Separate generation from post-processing in the pipeline graph. Route any video edit or extension calls to grok-imagine-video instead of assuming 1.5 supports them.
  5. Align prompt templates with the input mode: describe shot, lighting, camera movement, and audio for text-to-video; describe subsequent motion for image-to-video; and use reference images when consistency matters without locking frame zero.
  6. Keep budget math in the native unit of the target environment, tracking per-second USD plus image-input fees for the SpaceXSI API and rounded-up per-second credits for a hosted UI.

Where to Check Full Model Details

For the broader feature tour, launch timeline, and full credit table across durations, see the Grok Imagine 1.5 features and usage guide. For a similar pacing problem on another short-clip model, there is also an earlier piece on how to use the extra five seconds in a Gemini Omni 1.2 Flash clip. Tying model selection to verified endpoint constraints up front keeps storyboards and cost sheets aligned with what your runner will actually execute.

Top comments (0)