Grok Imagine (I2I) — AI Image Generator

About Grok Imagine (I2I)

Features

• Edits driven by a plain-language instruction; the source image can be a public URL, a base64-encoded data URI, or a stored file id from xAI's Files API.
• Multi-image editing accepts up to 3 source images in one request for combining subjects, transferring styles, and composing scenes.
• Multi-turn editing chains: each output can feed the next request for iterative refinement of a single image.
• Style transfer across a wide range of looks, from ultra-realistic photography to anime, oil painting, and pencil sketch.
• Single-image edits preserve the input's aspect ratio automatically; multi-image edits accept an explicit aspect ratio parameter.
• Results return as hosted URLs or base64 data with a per-image moderation flag.

What people use it for

• Retouching product or portrait photos with targeted, described changes instead of manual masking.
• Restyling existing images into illustration, sketch, or painting looks for creative reuse.
• Compositing subjects from up to 3 reference photos into one scene.
• Iterative art direction: start from a base image and refine details across several editing rounds.
• Converting brand photography into stylized campaign variants while keeping composition intact.

Reported strengths

• Three input paths (URL, base64, stored file id) make it easy to wire into existing asset pipelines.
• Multi-turn chaining is a documented first-class workflow, not a hack, which suits stepwise refinement.
• The quality tier's style range covers photorealism through hand-drawn looks in one model.
• Aspect-ratio preservation on single-image edits keeps outputs drop-in compatible with the original layout.

Reported limitations

• Single-image edits cannot change the output aspect ratio; reframing requires the multi-image path or a separate generation.
• Multi-image editing caps at 3 source images per request.
• The OpenAI SDK's images.edit() method does not work against this API (it sends multipart form data while xAI requires JSON), so ported OpenAI code needs rework.
• Edited results come back on temporary URLs and go through content-policy review before delivery.

Sources

1. Image Editing (xAI Docs)
2. Multi-Image Editing (xAI Docs)
3. grok-imagine-image-quality model page (xAI Docs)

What Grok Imagine (I2I) is good at

• +Committed style transfer: the photo-to-watercolor test produced a genuinely painted result — loose washes, illustrated linework, aged-paper tone — rather than the half-photographic hybrid some editors return.
• +Composition and identity survive the conversion: hat, jacket, harbor boats, and the subject's likeness all carried through the repaint.
• +Fast and cheap for an editing model: our edit finished in ~15 seconds at one of the lowest prices in the catalogue.

Where Grok Imagine (I2I) falls down

• −Faces flatten slightly in stylized conversions — the watercolor softened facial structure more than the source warranted; fine for illustration use, weaker for portrait-faithful edits.
• −Unlike its text-to-image sibling (six variants per run), the editing model returns a single output per generation, so exploration costs scale linearly.
• −Inherits the family's loose style adherence: expect an artistic interpretation of your instruction rather than a literal one.

Best for

• Fast, cheap style conversions
• Illustration looks from photos
• Iterating on a reference image

Sample outputs

Generated on Artifio with Grok Imagine (I2I). Each sample shows the exact prompt used.