I'll tell you what DALL-E is good at by telling you when I reach for it instead of Midjourney.
When a client says "I need the product in the foreground, on the left side, with the person's hand visible but not their face, and there should be a window in the background." When someone asks for "a vintage 1970s bus stop with the words 'Next Bus: Never' on the sign." When a brief calls for a specific scene with five specific elements in specific spatial relationships.
That's when DALL-E is the right tool. It listens. Midjourney interprets -- beautifully, often, but it interprets. DALL-E executes.
That distinction isn't a slight against either tool. They're optimized for different things, and understanding which to reach for is most of the skill.
What's New in DALL-E 3.5
OpenAI launched DALL-E 3.5 in late 2025, and the improvements over 3.0 are meaningful. Three areas show genuine progress:
Prompt adherence. Already DALL-E's strongest suit, it's better still. The model has improved spatial reasoning -- object placement, relative sizes, compositional relationships. I asked for "a coffee cup on the right side of the frame with a closed laptop on the left, both on a wooden desk, morning light from a window in the background." v3.5 nailed this on the first try. That sounds like a low bar, but try it with the other generators.
Text in images. This was DALL-E's Achilles heel and it's become a genuine strength. Phrases of 4-8 words render accurately and legibly most of the time. You still see errors on longer text and complex typography. But "Sale Ends Sunday" on a banner, or a brand name on a product label, now works without ten regeneration attempts. (Ideogram is still better specifically at text, but DALL-E's improvement here is real.)
Realism. Photorealistic human faces, hands, and skin tones have all improved. The earlier DALL-E versions had noticeable issues with faces -- slightly uncanny expressions, problematic hands. 3.5 isn't perfect but it's much better. Standard portrait photography now comes out usably realistic.
The ChatGPT Integration
Here's the practical advantage DALL-E has over every competitor: if you're already in ChatGPT, it's just there.
You're describing a project to ChatGPT. You can ask it to generate an image. You can have a conversation about what you want, let ChatGPT help you refine the concept, and then generate the image -- all in the same thread. The AI-to-AI assistance here is genuinely useful. "I need a hero image for this article about remote work. Here's what the article covers. Suggest an image concept and then make it."
The feedback loop is tighter than it is with Midjourney's interface. ChatGPT can explain what it's doing and why. You can give feedback in natural language. "The person looks too corporate -- make them look more like they're relaxed and working from a coffee shop." And it adjusts.
For people who aren't deep into AI image generation as a specialized skill -- which is most people -- this integration makes DALL-E the most accessible serious image generator available.
Prompt Adherence: Why It Matters
Let me be specific about this because it's the core case for DALL-E.
I ran the same set of 20 complex prompts through Midjourney v7 and DALL-E 3.5, rating each on how well it matched the specific requirements of the prompt. Midjourney scored 14/20. DALL-E scored 18/20. The prompts I used included spatial relationships ("on the left," "behind the foreground subject"), counts ("three candles"), color specifications ("navy blue jacket, not black"), and text ("the poster reads 'Opening Night, March 15'").
The difference was consistent. Midjourney would produce something beautiful that interpreted the spirit of the prompt. DALL-E would produce the prompt.
For advertising and commercial work, this often matters more than beauty. A client who says "the product must be in the foreground" doesn't want an artistic interpretation where it's tastefully in the background. DALL-E is the reliable tool.
Where DALL-E Falls Short
Artistic Quality
Straight comparison: Midjourney images feel more alive. They have more visual intention, more coherence of mood and light. DALL-E produces correct images. Midjourney produces felt images.
This gap has narrowed but hasn't closed. For editorial work, brand concepts, and anything where emotional resonance is the primary metric, Midjourney is still the better tool. DALL-E's outputs can look technically competent but aesthetically safe -- a sophisticated version of stock photography.
Volume Limitations
Included with ChatGPT Plus, you get 50 DALL-E generations per day. For a professional working on campaigns, that's not much. Midjourney on the Standard plan gives roughly 900 fast generations per month. For high-volume production work, the economics favor Midjourney -- or the API.
The OpenAI API pricing for DALL-E 3.5 HD quality runs about $0.080 per image at 1024x1024. For serious production volume, that adds up fast. A 200-image creative shoot through the API costs $16 -- not ruinous, but you're watching the meter.
Style Consistency
DALL-E doesn't have Midjourney's style reference or character reference features. You can describe a style in your prompt, but maintaining visual consistency across a set of images requires careful prompting on every generation. No shortcut to lock in "use this aesthetic."
For projects that require consistent visual branding across many images, this is a real limitation. Midjourney's --sref and --cref parameters are a better tool for that use case.
Inpainting Still Needs Work
DALL-E's edit feature -- the ability to select a region of an image and regenerate just that part -- exists and works. But Adobe Firefly's implementation is smoother and the output quality in edited regions is often better. For post-generation cleanup and adjustments, Firefly integrated into Photoshop remains the better tool.
DALL-E API: For Developers
This is an underappreciated angle. If you're building a product that needs image generation -- a custom marketing tool, a content creation workflow, an e-commerce product visualizer -- DALL-E's API is the most stable and well-documented image generation API available.
The prompt adherence advantage is even more valuable when you're generating images programmatically. You're not manually reviewing each output and prompting again -- your code is generating images based on data. You need the image to match the spec. DALL-E delivers that reliability more consistently than the alternatives.
OpenAI's API has strong uptime, consistent response times, and extensive documentation. For developers, this matters as much as image quality.
DALL-E vs. Midjourney: When to Use Which
I use both. Here's my actual decision tree:
Reach for DALL-E when:
- The brief has specific compositional requirements ("product on left, white background, shadow visible")
- You need text in the image (anything more than one word)
- You're working in ChatGPT and the conversation context is already established
- You need API access for programmatic generation
- Prompt accuracy matters more than aesthetic distinctiveness
Reach for Midjourney when:
- You're doing concept exploration and want to see unexpected but beautiful interpretations
- Mood and atmosphere are the primary goals
- You need character or style reference consistency
- You want the image to feel visually intentional and distinctive
- You're doing editorial, advertising, or brand mood work
Use both when: You're in a discovery phase. Run the same prompt through both. Midjourney shows you what's possible aesthetically. DALL-E shows you what your specific brief looks like executed. They're often telling you different things that are both useful.
For a full comparison of the major tools, see our Midjourney vs DALL-E vs Ideogram comparison. And if you want to get better results from any of these tools, our guide to writing AI image prompts covers the techniques that actually work.
Pricing
DALL-E 3.5 is accessible through two main paths:
ChatGPT Plus ($20/month): 50 DALL-E generations per day included. Also gets you GPT-4o access, code interpreter, and all other ChatGPT Plus features. If you're already using ChatGPT Plus for writing or research, DALL-E access is effectively free as a bonus.
OpenAI API: Pay per image. Standard quality at 1024x1024 runs about $0.040/image; HD quality is $0.080/image. Dall-E 3 at 1792x1024 widescreen is $0.120/image. For developers and high-volume professional users, the API is the practical access path.
No standalone DALL-E subscription exists -- you access it through ChatGPT or the API. That's fine for most users and a minor friction point for the few who want DALL-E access without ChatGPT.
Who Should Use DALL-E 3.5
DALL-E is the right choice if you:
- Already pay for ChatGPT Plus and want image generation included
- Need reliable prompt adherence for commercial or instructional work
- Are a developer building image generation into an application
- Need text rendered accurately in images
- Prefer working in a conversational interface where you can refine with natural language
Consider Midjourney instead if you:
- Do primarily editorial, advertising, or brand creative work
- Value aesthetic quality and mood over technical accuracy
- Need high-volume generation (Midjourney's Standard plan has better volume economics)
- Want character and style reference features for consistency
The Verdict: 4.1 out of 5
DALL-E 3.5 is the image generator I'd recommend to most people who don't already have a specialized workflow. The ChatGPT integration makes it the most accessible path to serious AI image generation. The prompt adherence is genuinely best-in-class. Text rendering has become usable.
The 0.9 deduction is for an artistic quality ceiling that Midjourney still clears more consistently, volume limitations on the ChatGPT Plus plan, and the lack of style/character reference features that make consistent branding harder.
If you're a developer or a non-designer professional who needs reliable image generation without learning Midjourney's prompting conventions, DALL-E 3.5 is the right tool. If you're doing creative work where images need to feel as much as they need to inform, learn Midjourney.
For most people, the correct answer is to start with DALL-E 3.5 in ChatGPT, see what it can do, and add Midjourney to your toolkit when you hit its limits.
Running into errors with DALL-E? Our DALL-E not working troubleshooting guide covers content blocks, greyed-out buttons, API errors, and access issues.
Top comments (0)