DALL·E 3 API playground
This playground calls the OpenAI Images API straight from your browser using your own API key, so you can test DALL·E 3 prompts and parameters without writing any code. Set the size, quality, and style, watch the cost estimate update live, then generate — the image and OpenAI’s revised prompt appear inline.
How it works
When you click generate, the tool sends a single HTTPS request directly to
https://api.openai.com/v1/images/generations with your prompt and options. The
key travels only in that request’s Authorization header; there is no server in
between. DALL·E 3 returns an image URL plus a revised_prompt — the version it
actually rendered after its automatic rewriting — which the playground shows so
you understand the output.
Understanding the three parameters
Size — DALL·E 3 supports three options: 1024×1024 (square), 1792×1024 (landscape), and 1024×1792 (portrait). Standard square is the cheapest; the wider landscape and portrait sizes cost more. Choose the aspect ratio that matches the intended use — a website banner wants landscape; a phone wallpaper wants portrait.
Quality — standard is the default and fast. hd runs a second generation
pass to improve fine detail and coherence in complex scenes — useful for anything
with text, intricate architecture, or many faces, but it doubles the cost for the
same size.
Style — vivid instructs the model toward hyper-real, dramatic imagery with
saturated colours; natural stays closer to a literal, photographic interpretation.
For product renders or documentary-style images, natural usually reads more
credibly; for fantasy or editorial art, vivid stands out.
The revised prompt — why it matters
DALL·E 3 always rewrites your prompt before rendering. The revision adds detail, clarifies ambiguity, and applies safety constraints. Seeing the revised version is genuinely useful: if the output is not what you expected, the revised prompt usually shows exactly where the model diverged from your intention — a word substituted, a detail added, or a subject repositioned. Use it to refine your next prompt rather than regenerating blindly.
Tips for better outputs
- Write prompts that describe the visual result, not an action: “a watercolour painting of a red barn at sunset” not “paint me a barn”.
- Specify camera angle, lighting, and medium explicitly — DALL·E treats these as distinct creative choices it will fill in itself if you leave them open.
- If you need text in the image, keep it short and add “exact text: [your words]” because DALL·E 3 still struggles with long strings.
- Always read the revised prompt before submitting a follow-up request; it saves a generation cycle when the fix is a simple phrasing tweak.
- Image URL expiry: the URL returned by the API is temporary. Download the image to disk immediately after generation; the URL will expire after a short window (typically about an hour) and cannot be regenerated without another API call and charge.