Creating images with ChatGPT — prompt and editing best practices
Key Points
- Use 1–3 clear sentences for prompts
- Iterate with small, targeted revisions
- Specify text and constraints explicitly
Summary
ChatGPT can generate and edit original images from plain-language prompts and multiple uploaded images. Use concise, clear prompts (typically 1–3 sentences) that explain the image's purpose, main subject, action, location, and desired visual style. For edits, be explicit about what to change and what must remain fixed, and refine images through small, targeted iterations.
Key Points
- Prompt structure: include purpose, subject, action, environment, and visual style; add framing, lighting, or material details when they matter.
- Keep prompts concise: clarity is more effective than length or clever phrasing.
- Use explicit constraints for fixed elements (e.g., “No logos or extra text” or “Change only X. Keep everything else exactly the same”).
- Iterate with targeted revisions: change one element at a time (e.g., “Make it brighter,” “simplify the background,” “tone down colors”).
- Multiple reference images: upload a small set, refer to images by number, and use clear spatial language (left/right/foreground/background) when combining elements.
- Text in images: put copy in quotes or ALL CAPS, specify font/size/color/placement, keep copy short, and spell uncommon brand names letter-by-letter for accuracy.
- Infographics and dense layouts: request "sharp text rendering" and plan to polish heavy text in design tools if needed.
- Additional considerations: obtain permission for likenesses, prefer generic/ownable designs over direct imitations, attribution is optional, and comply with organizational and OpenAI policies.
Quick checklist
- State purpose and key visual details up front
- Specify any absolute constraints (no logos, keep X unchanged)
- Apply one focused revision per iteration
- Use numbered image references and spatial language when combining uploads