Grok Imagine Image
What is Grok Imagine Image?
Grok Imagine Image is the speed-focused image generation route of xAI's Grok Imagine model. One endpoint covers both text-to-image and reference-guided edits: omit image_urls and the model generates from the prompt alone, or pass 1 to 5 reference URLs and the same model switches into edit mode anchored to those visuals. Output covers five aspect ratios (1:1, 2:3, 3:2, 9:16, 16:9). For quality-focused output, select the separate xai/grok-imagine-2.0-image model. The old enable_pro input is no longer supported.
Key features of Grok Imagine Image
Five features cover what you'll use in a real production loop.
Text-to-image and reference edit on one endpoint
Skip `image_urls` for fresh text-to-image. Pass 1 to 5 references and the same call switches into edit mode. No conditional routing in your client code, no second model id to manage.
Speed-focused iteration
Use `xai/grok-imagine-image` for quick drafts and rapid creative exploration. Quality is now selected with the separate Grok Imagine 2.0 Image model instead of a request toggle.
A separate quality model
Move final-quality runs to `xai/grok-imagine-2.0-image`. Both model IDs support text-to-image and reference editing, so adding references never changes your selected quality tier.
Up to 5 references per edit call
Edit mode takes 1 to 5 reference images to composite subjects, transfer a style, or build a scene. Anchor product geometry while rotating background or palette across runs, or carry character identity across a campaign.
Five aspect ratios that match Grok Imagine video
1:1, 2:3, 3:2, 9:16, and 16:9. Same set as the video model, so a still mocked at 9:16 maps cleanly to a vertical video without re-prompting per ratio.
Best for
Single endpoint, two workflows
Text-to-image and reference edit run on the same call. The presence of `image_urls` flips the workflow with no conditional routing in your client.
Two-track production runs
Draft quickly with `xai/grok-imagine-image`, then send the winning prompt to `xai/grok-imagine-2.0-image` for the quality pass.
Multi-reference colourway swaps
Up to 5 references per call lets you anchor product geometry while rotating background or palette across runs. Useful for catalogue spreads.
Fast concept exploration
Use the speed model to test compositions, palettes, and reference combinations before committing to a final-quality 2.0 render.
Cross-platform social sizing
The same prompt across 1:1, 9:16, and 16:9 generates a matched set without re-prompting per ratio. Pair with the Grok Imagine video model for consistent stills and videos.
NSFW workflows on permitted accounts
A single `enable_nsfw` flag handles policy gating with no extra integration. Useful for moderated adult-content pipelines that need one supported provider.
Model choice
Speed and quality are separate models. enable_pro has been removed.
Speed
Use xai/grok-imagine-image for brainstorming, bulk drafts, and rapid iteration.
Quality
Use xai/grok-imagine-2.0-image for Grok Imagine 2.0 quality. It accepts the same prompt, aspect-ratio, and reference-image inputs.
Text-to-image
Omit image_urls. The model generates from the prompt alone. Good for fresh concept work where no reference exists yet.
Reference-guided edit
Pass 1 to 5 references via image_urls. The call switches to edit mode, anchoring the result to those visuals, and returns a single image. Useful for product photography, character continuity, and background swaps.
Use cases
Run a brainstorming loop with xai/grok-imagine-image at 1:1 to pick a hero composition, then re-run the winner with xai/grok-imagine-2.0-image for the final delivery file. Build a colourway sheet by uploading a flat product photo as a reference; references activate edit mode without changing the chosen model. Anchor a campaign with up to 5 references per call so palette, geometry, and props stay consistent across a spread.
API examples
Call Grok Imagine Image from any language by POSTing to /v1/tasks. Full parameter docs live at docs.unifically.com/models/image/xai/grok-imagine-image.
curl -X POST https://api.unifically.com/v1/tasks \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"model": "xai/grok-imagine-image",
"input": {
"prompt": "A robot dancing in a field of sunflowers",
"aspect_ratio": "16:9"
}
}'
Successful submission returns a task_id. Poll GET /v1/tasks/<task_id> or set a callback_url on the request to receive the finished image URLs.
FAQs
People also ask
Grok Imagine Image is the image route of xAI's Grok Imagine model. One endpoint covers both text-to-image (when you omit references) and reference-guided edits (when you pass 1 to 5 reference images). Edit mode returns one image per call.
References are optional. Skip image_urls for pure text-to-image. Pass 1 to 5 URLs and the call switches to edit mode, with the model anchoring the result to those visuals. It composites subjects, transfers styles, and builds a scene from the references, so it fits product photography, character continuity, and background swaps.
Use the separate xai/grok-imagine-2.0-image model. The old enable_pro input has been removed; this model ID remains the speed-focused option.
For typography-heavy work where quality is the priority, select xai/grok-imagine-2.0-image. Use this speed model for rapid drafts and iteration.
Set upsample_prompt to true and the API rewrites your text prompt before sampling, expanding short or vague briefs into a fuller description. Disable it when you want the model to take your prompt verbatim.
Five. 1:1, 2:3, 3:2, 9:16, and 16:9. Same five ratios as Grok Imagine video, so the same composition can be mocked as an image and animated as a video without re-prompting per ratio.
Only on accounts with the policy enabled. Pass enable_nsfw: true on the call and, if the account is permitted, the result skips the safe-for-work filter. Without the flag, results stay safe-for-work.
Edit mode (when image_urls is set as input) returns a single image_url. Text-to-image mode returns one or more images as image_urls.
Guides and comparisons
Review pricing, limits, and tested outputs before running this model.
