DALL·E 3 in Bing and ChatGPT: A 2023 Walkthrough

I have just received access to DALL·E 3 in ChatGPT and want to compare it with Bing Image Creator. This walkthrough was published in October 2023: the access rules, prices, interface and limits discussed here describe that moment. Microsoft had [announced free DALL·E 3 access in Bing](https://blogs.bing.com/search/2023/10/DALL-E-3-now-available-in-Bing-Chat-and-Bing-com-create-for-free/) on October 3.

At the time, I could sign in with a Microsoft account and use Bing Image Creator. My interface showed 100 boosts, which I understood as faster generations, usually producing four images per prompt; after boosts ran out, I expected a longer wait. Even with boosts, the screen now says five minutes. That experience is not evidence of a permanent unlimited service.

In my ChatGPT Plus interface, I select the DALL·E 3 beta under GPT-4. The workflow differs from directly entering an image prompt: ChatGPT takes my request and writes more detailed prompts for the image model. I ask it for a logo for a company called Forward Future AI.

The system produces several different concepts, including a metallic emblem and a more dimensional rendering. Looking at the four outputs, I like one brain-like symbol, but several spell the name incorrectly. I can open a larger image, move between variations and download a result; I regenerate to see more options.

On the next attempt, ChatGPT expands the request into descriptive prompts including an upward arrow and digital circuitry. The designs are attractive to me, but the lettering still varies in accuracy. I copy one of those prompts to try in Bing and compare the result.

The Bing interface says a boost makes creation faster, but this request takes a long time. I also see a message saying a boost will not be used because the wait is so long. I suspect high demand; the video does not establish why the delay happened.

While waiting, I try the prompt in Leonardo.ai, which I like for its depth of features. The first four results arrive quickly but look similar and do not contain legible lettering. Switching to SDXL 0.9 produces more variety in my view, but the examples still lack useful text. A different fine-tuned logo model might behave differently; I do not test one here.

Next I ask ChatGPT for cute robot puppies playing in a field. It writes different prompts, which gives me a range of interpretations. I wonder how to make variations of a particular image and try copying its prompt back into the chat. Under high demand, the service returns only one image and says it could not generate all those requested.

When I paste a prompt back into ChatGPT, it rewrites it, leaving me unsure how to send the exact wording directly through that interface. I speculate that Bing and ChatGPT might share servers because both are slow, but I have not verified that. Microsoft and OpenAI are separate companies, so the coincident delays alone do not establish a shared infrastructure.

I also try the robot-puppy prompt in Leonardo.ai. The initial images are cute and cartoonish; after I request photorealism, some dogs look wrong to me. I prefer the DALL·E 3 examples in this small, subjective comparison, though the outputs and model settings differ.

The new Bing request is still not ready, so I review images I made earlier: cars, astronauts and black Labrador puppies in a garden. I like their detail and realistic appearance. A later ChatGPT robot-puppy result is more realistic than the earlier vector-style images, though I still see a painted quality and some odd repetitions.

Finally I request the words “Forward Future AI” in a colorful futuristic scene with an astronaut, and try the prompt in Leonardo.ai too. Some DALL·E 3 images spell the words correctly while others add or distort letters; the Leonardo examples shown also struggle with the text. For these examples, I find DALL·E 3 better at lettering, without treating the result as a comprehensive benchmark.

At the time of the video, I expect the ChatGPT Plus DALL·E 3 feature to continue rolling out. I would be interested in a deeper comparison with Leonardo.ai and Midjourney. The results here are a snapshot of the tools and capacity available in October 2023, not instructions or pricing guarantees for today.