How do I get a black-and-white archival look?▶
Append the fragment 'black and white archival photograph, 1930s documentary style, film grain, high contrast' to your subject. Use aspect_ratio 4:5 and the Default Image model (1 credit). Bake the constraint 'no text, no words, no logos' into the prompt itself — the default image model does not accept a separate negative_prompt parameter.
How do I get a photorealistic image?▶
Use 'photorealistic, dramatic natural light, shallow depth of field' as the style fragment with aspect_ratio 16:9 and the Default Image model. This is the most common hero-image recipe and costs 1 credit.
How do I make a cartoon or illustrated version?▶
Append 'playful cartoon illustration, bold outlines, flat vibrant colors'. Default Image model, 1 credit, aspect_ratio 16:9 for banners or 4:5 for portraits.
How do I get a watercolor look?▶
Use the fragment 'soft watercolor painting, wet on wet, paper texture, gentle washes, painterly'. Avoid adding 'illustration' — the painterly quality comes from explicit paper/wash cues.
How do I render a flat vector or minimalist look?▶
Use 'flat vector illustration, minimalist, clean geometric shapes, limited palette, modern poster style'. Best for SaaS landing pages and dashboard art.
How do I get a 3D isometric look?▶
Append 'isometric 3D render, soft studio lighting, low-poly, clay material, cozy diorama'. The 'low-poly clay' qualifier keeps the result lightweight instead of photoreal 3D.
How do I make a cinematic film still?▶
Use 'cinematic film still, anamorphic lens, teal and orange color grade, atmospheric haze, wide shot' with aspect_ratio 21:9. Great for article heroes and video thumbnails.
How do I get a vintage travel-poster feel?▶
Use 'vintage travel poster, 1950s offset print, bold flat color blocks, screen-print halftone, heroic composition' with aspect_ratio 2:3.
Why can't I just use a negative_prompt parameter?▶
On the Default Image model, a separate negative_prompt field is accepted at submit but causes the job to fail at generation. Always bake exclusions into the prompt text itself ('no text, no words, no logos'). The Premium Image model (gpt-image-2, 4 credits) does support negative_prompt natively if you need it.
Which model should I use for text-in-image or strict prompt adherence?▶
Use the Premium Image model (4 credits) — it has stronger prompt adherence and renders in-image text. The Default Image model (1 credit) is faster and cheaper for everything else. See GET /api/models for the live catalog.