Current status: DALL·E 2 is a legacy OpenAI image model. OpenAI’s current documentation says it has been deprecated and removed from the API, with GPT Image 2 recommended for current image generation and editing. If you landed here looking for a DALL·E 2 tutorial, the practical path is now to understand what changed and move your workflow to a supported GPT Image model.
This updated guide preserves the history of DALL·E 2 while explaining the current OpenAI image-generation workflow, how the Image API differs from the Responses API, and what developers should check before migrating an old integration.
Is DALL·E 2 Still Available?
No—not as a current OpenAI API model. The official DALL·E 2 model page labels it deprecated and states that it has been removed from the API. OpenAI recommends GPT Image 2 for current generation and editing.
That means an older tutorial showing a request with model: "dall-e-2" should not be treated as a working setup guide. Archived examples can still be useful for historical context, but new projects should use a supported model and the current API documentation.
DALL·E 2 vs. GPT Image 2
| Area | DALL·E 2 | GPT Image 2 |
|---|---|---|
| Status | Deprecated and removed from the OpenAI API | Current GPT Image model recommended in OpenAI documentation |
| Primary role | Legacy text-to-image generation, edits, and variations | Current image generation and editing |
| Image input | Legacy workflow and constraints | Supports image input and output for high-fidelity editing workflows |
| API path | Old integrations may reference the Images API | Can be used through the Image API; conversational image workflows can also use the Responses API tool |
| New project choice | Do not build a new dependency on it | Check current model availability and use the supported documentation |
The model name is only one part of the migration. Output encoding, supported parameters, image sizes, quality controls, moderation behavior, cost calculation, and edit inputs may differ. Test the complete request and downstream image-processing workflow.
Two Current Ways to Generate OpenAI Images
The official OpenAI image generation guide describes two main API approaches.
Image API
Use the Image API when your application needs a direct generation or edit operation. For a straightforward “prompt in, image out” workflow, this is usually the simpler choice. With a current GPT Image model, the API can generate a new image or modify an existing image using a prompt.
Responses API With Image Generation
Use the Responses API when image generation is one step inside a conversation or multi-step flow. OpenAI’s documentation highlights multi-turn editing and flexible contextual inputs as advantages. The mainline model can call the image-generation tool as part of a broader response workflow.
| Choose the Image API when… | Choose the Responses API when… |
|---|---|
| You need a direct generation or edit request. | You need conversational or iterative image work. |
| Your application already handles the surrounding workflow. | The image should be generated as part of a larger agent flow. |
| You want to select a supported GPT Image model directly. | You want model context and tool use within the same response process. |
Basic GPT Image 2 Generation Workflow
1. Confirm Current Access and Requirements
Check the current model page, account access, organization requirements, rate limits, and pricing before building. Availability and costs can change, so avoid copying old values from a DALL·E 2 tutorial.
2. Write a Production-Oriented Prompt
Describe the subject, setting, composition, lighting, style, required text, and intended use. For edits, say what should change and what should remain unchanged. Concrete constraints produce a request that is easier to test than a long list of generic quality adjectives.
Example structure: “Create a landscape editorial illustration of [subject] in [setting], with [composition] and [lighting]. Keep the left third simple for a headline. Do not include logos or watermarks.”
3. Choose Output Settings
Select a supported size, quality level, and format based on the current model documentation. Larger or higher-quality outputs can require more processing and cost. Generate only what the project needs, then test the output in the real publishing layout.
4. Decode and Store the Result Securely
Current GPT Image examples return image data that your application decodes and saves. Validate the response, handle errors, use safe filenames, and store files according to your application’s privacy and retention requirements.
5. Review the Image Before Use
Inspect hands, faces, logos, text, small objects, product geometry, and factual or branded details. Generation can produce attractive but incorrect visuals, so a human approval step remains important.
Migration Checklist for an Old DALL·E 2 Integration
- Replace the deprecated model with a currently supported GPT Image model.
- Read the current model and image-generation documentation instead of relying on archived sample code.
- Verify the endpoint and request body used by your chosen API.
- Update supported parameters for size, quality, number of outputs, format, and editing inputs.
- Handle the current response format correctly in your application.
- Recalculate costs using the live pricing documentation.
- Review rate limits and any organization-verification requirements.
- Retest moderation and error-handling paths.
- Run visual regression checks for prompts your product uses frequently.
- Remove user-interface copy that still promises DALL·E 2 specifically.
How to Write Better Image Prompts
- Be explicit about the deliverable: photo, illustration, icon, poster, mockup, or diagram-like visual.
- Specify composition: camera distance, angle, subject placement, and negative space.
- Describe lighting and environment: these often affect the result more than generic “high quality” wording.
- Include exact constraints: aspect orientation, background, required elements, and prohibited elements.
- For editing, separate change from preservation: state what to replace and what must remain stable.
- Iterate deliberately: change one major instruction at a time and compare results.
If you prefer a no-code, conversational starting point, see our beginner guide to using ChatGPT. For alternative creative workflows, compare our Midjourney AI tutorial and Canva AI image generator guide.
Frequently Asked Questions
Why was my DALL·E 2 API request rejected?
The model has been deprecated and removed from the API. Update the integration to a supported model and follow the current image-generation guide rather than changing only a minor request parameter.
What should replace DALL·E 2?
OpenAI’s current DALL·E 2 model page recommends GPT Image 2 for image generation and editing. Confirm that recommendation against the live documentation when you implement it.
Can I keep an old article or product page about DALL·E 2?
Yes, when it clearly identifies the model as legacy, answers the current-status question immediately, and directs readers to a supported alternative. Do not present archived instructions as a working current tutorial.
Final Takeaway
DALL·E 2 matters historically, but it is no longer the right foundation for a new OpenAI image workflow. Keep legacy references for context, move current generation and editing to a supported GPT Image model, and test the entire integration—not only the model-name field.