DALL·E Prompt Builder

GPT Image 2 (the model now behind ChatGPT's image generation, formerly DALL·E 3) doesn't take parameters — it takes a brief. Inside ChatGPT your words are rewritten into a detailed internal prompt, so the winning move is a complete, specific sentence rather than keyword soup. This builder assembles exactly that: orientation phrased in words (there is no --ar here), your subject as the core clause, and style, lighting and palette folded into natural language, with the anti-garbled-text clause appended.

Clarifier: ChatGPT now generates images with GPT Image 2. OpenAI retired DALL·E 3 in May 2026, but the brief-writing rules below are exactly the same — paste the assembled prompt into ChatGPT and refine conversationally.

Fill in a subject above — the prompt assembles here as you choose.

How to use it well

Paste the assembled brief into ChatGPT, then refine conversationally — "same image but at dusk", "make the pot terracotta" — instead of rebuilding from scratch; iteration is GPT Image 2's real superpower. When a scene needs several distinct elements, enumerate them explicitly ("exactly four icons: chat, camera, music, mail") because GPT Image 2 follows checklists more reliably than any model we test. And keep "no text anywhere in the image" unless you genuinely want lettering — it's the single most common fix in our DALL·E prompting guide.

For known-good starting points, the DALL·E library publishes every tested brief with its result — the sticker packs and app icons hubs are where ChatGPT's image generator earns its keep.

Frequently asked questions

Does DALL·E use --ar or other parameters?

No. GPT Image 2 (the model behind ChatGPT image generation since DALL·E 3 was retired in May 2026) takes no parameters at all — it takes a brief. There is no --ar; you phrase orientation in words instead, and it renders only three shapes: square, wide 16:9 and tall. Inside ChatGPT your words are rewritten into a detailed internal prompt, so a complete, specific sentence beats keyword soup every time.

How do I iterate on a DALL·E image?

Refine conversationally rather than rebuilding from scratch — "same image but at dusk", "make the pot terracotta". Iteration is the ChatGPT image generator's real superpower; each message adjusts the previous result. There are no seeds, so you cannot reproduce an image exactly, but the back-and-forth is fast. Paste the builder's assembled brief into ChatGPT first, then steer it one change at a time.

How do I stop DALL·E from adding garbled text to images?

Keep the clause "no text anywhere in the image" in your brief unless you genuinely want lettering — it is the single most common fix we apply. When a scene needs several distinct elements, enumerate them explicitly, such as "exactly four icons: chat, camera, music, mail", because GPT Image 2 follows checklists more reliably than any other model we test. The builder appends the anti-garbled-text clause for you.