GPT IMAGE — ONBOARDING & EMPTY STATES▌
First-run flows with illustration-lite accents.
Prompt Generator
Generate Your Prompt
Fill in the fields below to generate an optimized prompt based on best practices
Following OpenAI text-in-image quoting guidance.
Examples
Filled example
Input Variables
Frequently Asked Questions
Best Practices
Critical Best Practices
Ordered structure (scene → subject → constraints)
criticalWrite prompts in consistent order: background/scene → subject → details → exclusions. Include intended use (ad, UI mock, infographic) so the model sets polish level.
Example:
"Create a realistic mobile UI mock… White background … place in iPhone frame. No decorative clutter."
Lock in-image typography
criticalPut verbatim copy in quotes or ALL CAPS; specify placement, contrast, font style. For dense layouts use higher quality tier. Spell tricky brands letter-by-letter if needed.
Recommended Best Practices
Use explicit photorealism cues
recommendedFor believable photography, include “photorealistic” or “real photograph” plus lens, framing, lighting, and texture (fabric wear, pores). Avoid sterile studio-perfect language unless intentional.
Edits: “change only X; keep…”
recommendedFor image→image edits repeat invariants each turn: identity, layout, logos, angles. Say ‘do not alter saturation/lighting/objects’ when drift is unacceptable.
Optional Enhancements
Quality vs latency
optionalgpt-image-2 defaults for new work; use lower quality tiers for drafts / volume. Prefer medium/high for dense text, infographics, fine faces, slides. Very large canvases (>2K) may be less consistent.