xAI publishes no prompting guide for Grok Imagine. What it publishes is a control surface — aspect ratio, resolution, quality, duration, references — and a handful of its own example prompts. This sheet is those, quoted, plus the shape the examples share. What it costs against Veo and Gemini Omni is on the video model comparison.
- xAI’s own prompts are short. Subject plus style, one line: “A collage of London landmarks in a stenciled street‑art style”.
- Most of the control is in parameters, not words. Ratio, resolution and quality are fields, so do not describe them in the prompt.
- No negative prompt is documented — the parameter list does not mention one either way.
- Editing is a chain. xAI: “Chain multiple edits together by using each output as the input for the next.”
- Text-to-video is really image-to-video. The model makes a first frame from your prompt, then animates it.
- Video defaults to 480p. If you want HD you have to ask for it.
Every rule and every example prompt on this sheet is quoted from xAI’s own Imagine API documentation, read at source on 16 September 2026. Where a line is this site’s reading rather than xAI’s wording, it says so.
The formula
xAI publishes no formula. This shape is read off its own example prompts in Image Generation and Video Generation, read at source 16 Sep 2026.
Template:
[subject, concrete] [in a named style]. parameters: aspect_ratio · resolution · quality · (video) duration
xAI’s own examples, verbatim:
A collage of London landmarks in a stenciled street‑art style
A futuristic city skyline at night
Render this as a pencil sketch with detailed shading
Make the water crash down and slowly pan out the camera
Two of those are generation, two are editing and animation. None of them is longer than a sentence, and none of them describes a camera lens, a resolution or an aspect ratio — because those are fields.
aspect_ratio — not the word “vertical”resolution — video is 480p unless you askquality — auto means low on generationduration, and n for variationsBreakdown — what each part does
Subject and style
Style is the lever xAI names. It says grok-imagine-image-2.0 “supports a wide range of visual styles, from ultra-realistic photography to anime, oil paintings, and pencil sketches”, and that you get them by “describing the desired aesthetic in your prompt”. Name the style; do not imply it.
If your code still calls grok-imagine-image-quality, switch it to grok-imagine-image-2.0: xAI retires the older name on 2 November 2026. Old code will not break: after that date the old name is served by 2.0 at low quality, so switch to control the setting yourself.xAI, migration guide, read at source 22 Sep 2026: “After November 2, 2026, requests to grok-imagine-image-quality are served by grok-imagine-image-2.0 with quality set to low. The slug continues to resolve, so you do not need to change your code to avoid breakage.” · xAI, Imagine Image Quality Retirement, read at source 18 Sep 2026: “Effective November 2, 2026, the grok-imagine-image-quality model slug is retired from the xAI API.”
Aspect ratio is a parameter
There are fourteen documented ratios, from 1:1 to 21:9, plus auto. xAI: when omitted, the default is auto, “which lets the model pick the best ratio for the prompt”. Writing the word vertical into the prompt competes with a field that already does the job.
Resolution and quality
Images are 1k by default, with 2k available. Quality takes low, medium or auto, and xAI is specific about what auto means: “Auto currently uses low for image generation and medium for image editing”. It also warns about the bill: “Images are billed at the quality they are served at.”
Variations come from n, not from re-prompting
The n parameter takes 1 to 10 in a single request. xAI’s own advice for variations on one prompt is sample_batch() with n, because “that generates all images in a single request and is the most efficient approach for same-prompt generation.”
Editing is a chain, not a rewrite
Give it a source image and describe the change. xAI: “Chain multiple edits together by using each output as the input for the next. This enables iterative refinement; start with a base image and progressively add details, adjust styles, or make corrections.” Up to five source images can go into one edit.
Video starts as a still
Worth knowing before you write a motion prompt. On text-to-video, xAI says the model “generates a first frame from your prompt, then animates it”, and “the intermediate image is not returned”. So the first half of the prompt is doing image work and the second half is doing motion work — which is exactly the shape of xAI’s own example: a subject, then “slowly pan out the camera”.
The video limits, in one place
| Setting | What the docs say |
|---|---|
| Duration | 1–15 seconds |
| Default resolution | 480p |
| 1080p | text-to-video and image-to-video, on video-1.5 |
| Reference-to-video | capped at 720p |
| Video editing | keeps the input’s duration, capped at 8.7s |
| Default ratio | 16:9 |
| Audio | on by default |
xAI, Video Generation, read at source 16 Sep 2026: “The allowed range is 1–15 seconds”; “Generated videos include an audio track by default”; “The edited video retains the duration of the original, which is capped at 8.7 seconds.”
Asking for a long clip is a parameter change. Asking for a good one is a prompt change. Do not confuse the two — the defaults here are low, and silence about them costs you resolution.
Does it take a negative prompt?
Read across xAI’s Image Generation, Image Editing and Video Generation pages on 16 Sep 2026: the documented parameters are n, aspect_ratio, resolution, quality, response_format, duration and the reference fields. None is a negative prompt, and no page states that negatives are unsupported.
Absence from a parameter list is not a vendor statement. Treat it as unknown, not as a documented no — and write exclusions as description, which is what every other image model in this collection now asks for.
Run this before you send
- Is the style named? xAI’s lever is the aesthetic word, not the adjective pile.
- Is the ratio a field? If it is in the prompt, move it.
- Did you set quality? Auto means
lowon generation, and you pay for what you get served. - Do you want variations? Use
n, not ten requests. - Is this a video? Then the first clause is a still and the second is the motion.
- Did you ask for a resolution? Video is 480p unless you say otherwise.
- Is the URL saved? xAI: “URLs are temporary, so download or process promptly.”
Free or paid? What the free plan allows, who owns the output, and when you need to pay: free vs paid AI image, video and music tools.
xAI, Imagine Overview, Image Generation, Image Editing and Video Generation, all read at source on 16 September 2026.
API surfaces change faster than sheets do. Re-read before you build on this.