DALL-E GPT Retired: ChatGPT Images and Local Alternatives

DALL-E 3 was OpenAI's cloud-only image generator. The DALL-E 3 API was retired on May 12, 2026. It never ran locally.

At a Glance

TL;DR - Is It Worth It?

ChatGPT Plus still includes hosted image generation, but that is no longer DALL-E 3. OpenAI retired the DALL-E 3 API on May 12, 2026. Current OpenAI Image API docs point to GPT Image models, including gpt-image-2. ChatGPT image generation remains cloud-only. Local Stable Diffusion is the offline alternative.

Top 5 Features

1. ChatGPT integration - you describe what you want in a conversation, and ChatGPT rewrites your prompt behind the scenes to get better results. No prompt engineering required.
2. Strong prompt adherence - DALL-E 3 actually follows complex, multi-part instructions more faithfully than most competitors. In practice, it's one of the few tools where "a red car parked next to a blue bicycle" reliably gives you both objects.
3. Text rendering in images - it can generate legible text inside images (signs, labels, logos). Not perfect, but better than Midjourney or most Stable Diffusion setups.
4. HD and style options - the API lets you pick "standard" or "hd" quality, and "vivid" or "natural" style. HD adds ~10 seconds but visibly improves detail.
5. Developer API (historical) - DALL-E 3 had a REST API. That model ID no longer works after the May 12, 2026 retirement. New integrations should use OpenAI's current GPT Image models or a local stack.

Requirements & Setup

There's nothing to install. Sign into ChatGPT or create an OpenAI API account. For ChatGPT, you type a prompt and get an image. For the API, you send a POST request with your prompt and receive a URL or base64 image. The barrier to entry is genuinely low - which is DALL-E's actual competitive advantage over everything else.

Pricing Breakdown

The DALL-E 3 API was retired on May 12, 2026, so the API price rows below are historical. ChatGPT image generation is still a hosted OpenAI product on GPT Image models, not a local install. The marketing previously said ChatGPT Plus included image generation. In practice, there are rate limits - 50 images per 3-hour window - and during peak hours, generation slows noticeably. The API is more consistent but adds up fast: 100 HD images costs $8. OpenAI's newer GPT Image 1 Mini starts at $0.005/image, which makes DALL-E 3's API pricing look dated.

Limitations

• DALL-E 3 API is retired - OpenAI announced November 14, 2025 that DALL-E 3 would leave the API on May 12, 2026. That date has passed. Current OpenAI Image API docs name GPT Image models, including gpt-image-2. Azure retired DALL-E 3 earlier, on March 4, 2026.
• Content filters are the strictest in the market - DALL-E refuses to generate public figures, anything resembling violence, and often blocks prompts that are entirely SFW. Users report that even neutral historical or fantasy imagery gets flagged. "Prove it" - browse any Reddit thread on DALL-E censorship and you'll find hundreds of examples.
• No fine-grained control - you can't adjust composition, lighting, camera angle, or pose with any precision. There's no ControlNet, no LoRA support, no inpainting worth mentioning. If you need creative control, this isn't your tool.
• Quality has plateaued - the LM Arena leaderboard (December 2025) ranks DALL-E 3 at 10th place with GPT Image 1.5 at #1 (score 1264). FLUX models from Black Forest Labs now dominate professional image work. DALL-E 3's outputs often have a distinct "AI sheen" - slightly too smooth, slightly too lit.

How It Compares

Bottom Line

Use hosted OpenAI image generation if you:

• Already pay for ChatGPT Plus - hosted image generation is bundled, but it is not DALL-E 3 anymore
• Need fast, no-setup image generation - nothing matches the simplicity of typing a prompt into ChatGPT
• Build apps with OpenAI's current Image API - that path is hosted GPT Image models, not DALL-E 3

Skip DALL-E 3 if you:

• Want creative control - ComfyUI or Stable Diffusion with ControlNet gives you actual precision
• Need uncensored generation - local tools like Fooocus or Stable Diffusion don't have content filters
• Need a local/offline path - DALL-E never ran on your GPU; Forge, ComfyUI, and LocalForge AI do

For anyone exploring local alternatives, tools like ComfyUI, Fooocus, and LocalForge AI can run Stable Diffusion and FLUX models on your own hardware - no subscription, no content filters, no deprecation notices.