DALL-E 3: My Take After Two Years in ChatGPT and the OpenAI API

In short: DALL-E 3 is OpenAI's text-to-image model, accessible through ChatGPT, the OpenAI API, and Microsoft Designer or Copilot. Its main strengths are understanding natural-language prompts and embedding legible text in images. Users own and can commercially use generated images, though strict content filters and limited stylistic variety remain its key drawbacks.

I folded DALL-E 3 into my workflow the moment it went public in late 2023. Since then I've had to generate several thousand images — article illustrations, client visuals, mockups, training materials. This article is my field report condensed: what DALL-E 3 genuinely does well, where it falls short, and the cases where I still reach for Midjourney or FLUX instead.

What it actually is

DALL-E 3 is OpenAI's text-to-image generation model. I mostly use it through three doors depending on the context. Inside ChatGPT (account-based plans) for natural conversation with inline generation — that's the fastest way to iterate while you talk it through. Through the OpenAI API when I'm wiring it into a script or a client platform. And through Microsoft Designer or Copilot when I'm working inside a Microsoft 365 environment — Microsoft is OpenAI's strategic partner and exposes DALL-E 3 across several products, sometimes for open access.

The real strength: prompt understanding

This is what flipped me onto DALL-E 3 back then. Where Midjourney demands a technical syntax (parameters, ratios, precise stylistic keywords), DALL-E 3 responds well to natural, descriptive language. I can write in plain English: "a modern Scandinavian kitchen with a large skylight, a light-wood central island, a snow-covered garden visible in the background, warm atmosphere, realistic photo" — and get a coherent result on the first try nine times out of ten.

That fine-grained interpretation is precious for three audiences I work with: beginners who never learned the Midjourney syntax, writers who want complex narrative illustrations with several elements in a single scene, and designers in brainstorming mode who need to test an intention fast without calibrating every parameter.

Text in the image: a clear leap

DALL-E 3 is one of the most advanced models for embedding legible text inside images. I use it for visuals with slogans, packaging mockups, posters, and illustrations with text-based logos. The fidelity is still imperfect — some letters can still come out distorted, especially on long words — but the qualitative leap over earlier generations is real. For a quick mockup to put in front of a client, it's more than good enough.

How I access it depending on the context

Several doors, depending on the use case. ChatGPT Plus or Pro for reasonable unlimited access bundled into the access plan — that's my daily reflex. The OpenAI API for volume usage, with per-image billing and several resolutions available — that's what I use for my automated illustration pipelines. Microsoft Copilot or Designer for open access with quotas — handy when I'm working for a client on Microsoft tooling. And Bing Image Creator, open access with daily quotas — perfect for letting a client discover DALL-E 3 before they decide to invest.

For professional volume usage, the API stays the most predictable option. For steady individual use, ChatGPT Plus offers the best simplicity-to-cost ratio.

Usage rights: what OpenAI says

OpenAI clarifies in its usage policies that users own the rights to the images they generate and can use them commercially, provided they respect the usage policies (no illegal content, no impersonation of real non-consenting people, no protected trademarks without authorization). In practice, I use DALL-E 3 with no issue for commercial illustration of blogs and social media, marketing visuals, professional presentations, and e-commerce visuals for ambiance or fictional products.

Be careful about the legal status of the works themselves: in the United States, the Copyright Office has confirmed that an image generated purely by AI without significant human creative input is not protectable. For clients who want to protect their visuals, I recommend a round of human retouching after generation.

DALL-E 3 vs Midjourney vs FLUX

My three main image tools sit in different places, and I use them for distinct cases. DALL-E 3 for prompt fidelity, native ChatGPT integration, and text handling. Midjourney for pure aesthetic quality, fine artistic control, and the prompt community. FLUX (Black Forest Labs) — which I cover in detail in my article on Stable Diffusion — for open source, on-premise deployment, and full control of the data.

For a precise narrative visual with a complex scene, DALL-E 3 is often the most efficient. For an agency-grade aesthetic render, Midjourney keeps the edge. For a privacy-first integration at a sensitive client, I turn to FLUX running locally.

The limits I've observed

A few things to watch that I've noted across projects. The content filters are strict — some perfectly legitimate requests get refused (political subjects, celebrities, brands). The style is more "polished" and recognizable, with less variety than Midjourney when you're after a singular artistic render. There's no native "variations" feature as fluid as Midjourney's. The per-image API access conditions isn't trivial for heavy usage. And there's no specific private mode: for confidential work through ChatGPT, I systematically check the account settings to disable contribution to model improvement.

My typical use cases

I pull out DALL-E 3 for: article illustrations with a precise mood, presentations to replace generic stock photos, quick visual posts for social media, visual brainstorming of a marketing concept across a few iterations, storytelling for children's educational materials (with some consistency limits between images), and product mockups for packaging or posters.

My read for Trust-Vault

On my Trust Score methodology, DALL-E 3 and OpenAI show several solid strengths. Reliability: recognized render quality, especially on complex prompts. Transparency: public access conditions, very complete API documentation, a published model card. Security: SOC 2 Type II, Enterprise options with additional guarantees. Privacy: on ChatGPT Plus, data can be used for improvement unless you opt out — I recommend enabling it. On the API and the Enterprise offering, strict commitments apply by default.

For most professional use cases, DALL-E 3 through ChatGPT is a pragmatic choice. For strong confidentiality constraints, I look at the Enterprise offering or on-premise alternatives like FLUX or Stable Diffusion.

--- Sources: OpenAI Usage Policies; OpenAI API documentation; US Copyright Office — Policy on AI-generated works 2023; Black Forest Labs FLUX.1 release notes; Microsoft Designer documentation.

Further reading

For a complementary implementation angle, read AI Tools for SMEs: The Stack I Actually Deploy in 2026.

For a complementary implementation angle, read Prompt Engineering: The Techniques I Actually Use Daily.