Image Generation

Generate ad-ready images from text prompts, with optional style presets and brand reference images.

Models

Model	Description
Grok Image 2.0	Detailed, general-purpose images (default). Supports the Quality setting
Grok Image	Fastest Grok images
GPT Image 2.5 Sunburst (precise)	OpenAI model that follows detailed briefs closely: exact colours, counts, layouts and character sheets. About 10 to 15 seconds at Medium
GPT Image 2.5 Flare (fast)	Faster OpenAI model for quick drafts, less precise on small details
Luma Photon 1	Luma's image model
Luma Photon Flash	Faster Luma image model

The GPT Image 2.5 models appear where your Ads Optimiser deployment offers them. Every image uses one image generation from your plan, whichever model makes it.

When to choose Sunburst

Grok suits open-ended scenes. Choose GPT Image 2.5 Sunburst when the image has to match a detailed brief, for example:

• Exact colours or counts: "an overshirt with exactly four buttons, top to bottom: grass green, teal, amber yellow, coral pink"
• Character sheets: the same person from set angles, a full-body shot and a close-up. Make each view a separate image rather than one composite
• Thumbnails with layout space: "face on the right, the upper third left empty for a title"

See Create an AI influencer for a worked example.

Automatic fallback to Sunburst

If a Grok image fails on the provider's side (a server error or a timeout), Ads Optimiser retries it once on GPT Image 2.5 Sunburst, with the same prompt and reference images. The job page then says so, for example:

• Generated with GPT Image 2.5 Sunburst after Grok failed (500).
• Grok timed out; retried once on GPT Image 2.5 Sunburst. (shown when the retry failed too)

The job's model is then the model that actually made the image. The retry still counts as one generation. Requests the provider refuses, such as a moderation block, are never retried.

Settings

Quality

Grok Image 2.0:

Quality	Description
Low	Faster, around 15 seconds (default)
Medium	Highest detail, around 50 to 70 seconds
Auto	The model decides: low for new images, medium when editing with reference images

GPT Image 2.5 Sunburst and Flare:

Quality	Description
Low	Cheapest
Medium	Recommended (default), about 10 to 15 seconds
High	Most detail, costs more
Auto	The model decides

Grok Image and the Luma models have no quality setting.

Example: same prompt, different settings

These three images use the same prompt, a 9:16 aspect ratio and 1K resolution. Only the model and quality change.

Ultra-realistic product shot of a sleek matte black cold brew coffee can glistening with condensation, resting on warm golden sand at golden hour. Palm fronds and turquoise waves softly blurred in the background. Warm cinematic sunlight casts long gentle shadows and a radiant rim light along the can's edge. Low angle, the can is the hero against an expansive horizon. Sun-kissed colour grading with teal and coral tones. No text or logos on the can.

Grok Image 2.0, Low About 14 seconds	Grok Image 2.0, Medium About 70 seconds	Grok Image About 6 seconds

Low is a good default for drafts and variants. Switch to Medium for the final image when fine detail matters, such as condensation, textures or product edges.

Example: a detailed character brief

These images come from the same character description: a woman with a jet-black bob wearing a pale-sage overshirt with exactly four buttons, top to bottom: grass green, teal, amber yellow, coral pink.

GPT Image 2.5 Sunburst, Medium Four buttons, right order	GPT Image 2.5 Flare, Medium Close, but the green is lime	Zoomed in Left: Grok full-body shot with extra buttons. Right: Sunburst edit

Aspect ratio

Ratio	Use case
9:16	TikTok, Instagram Stories, Reels (recommended)
16:9	YouTube thumbnails, landscape ads
1:1	Social media posts, profile images
4:3 / 3:4	Presentations, portraits
3:2 / 2:3	Photography-style
2:1 / 1:2	Banners
21:9 / 5:2	Ultrawide and wide banners (Grok models)

GPT Image 2.5 models deliver three sizes only:

You choose	You get
1:1	Square, 1024 x 1024
2:3 or 9:16	Portrait, 1024 x 1536 (9:16 is delivered as 2:3)
3:2 or 16:9	Landscape, 1536 x 1024 (16:9 is delivered as 3:2)
Auto	The model chooses

Crop a 2:3 image to 9:16 if you need an exact TikTok frame, or keep important content away from the top and bottom edges.

Resolution

Resolution	Description
2K	Highest quality (recommended for ads)
1K	Standard quality

Resolution applies to the Grok models. The GPT Image 2.5 and Luma models size the image from the aspect ratio alone.

Style presets

Style presets append a style description to your prompt automatically:

• Cinematic: cinematic lighting, film grain, dramatic shadows, shallow depth of field
• Minimalist: clean aesthetic, negative space, muted tones, modern design
• Vibrant / Pop: saturated colours, bold contrast, energetic composition
• Professional: corporate look, studio lighting, polished finish
• Retro / Vintage: warm colour grading, nostalgic 70s–80s film look
• Neon / Cyberpunk: neon glow, dark background, vivid pink and blue accents
• Natural / Organic: soft daylight, earth tones, authentic lifestyle feel
• Luxury / Premium: rich dark tones, gold accents, elegant composition
• Product Photography: clean background, studio lighting, sharp product detail
• Lifestyle: candid natural moments, real-world setting, warm ambient lighting

Reference images

Add up to 5 reference images (4 with Luma models) to guide the AI. The GPT Image 2.5 models take up to 5, like Grok. The model blends your brand elements (logo, colours, products) into the generated image.

• Select from the media library (previously uploaded or generated images)
• Paste an external URL
• Without references, a new image is generated from scratch

Retry

If an image job fails or stalls, tap Retry on the job detail page to re-run generation with the same parameters.