Abstract

• ChatGPT’s image-generation capabilities create unintentionally hilarious cartoons with mangled textual content.
• The brand new GPT-4o mannequin is significantly better at producing photographs with clear textual content however is not used for scheduled duties.
• Lack of updates to scheduled duties means the dream of good cartoons continues to be on maintain.

Individuals are going nuts for ChatGPT’s new image-generation capabilities, creating all the pieces from photographs of themselves within the type of Studio Ghibli to pictures of different individuals within the type of Studio Ghibli. Extremely, ChatGPT may even make photographs in different types, too.

Sarcasm apart, I used to be very excited when the brand new options had been introduced, as I assumed it could imply an improve in high quality for my favourite scheduled activity. Sadly, I used to be incorrect.

ChatGPT Makes a Cartoon Each Morning (and They’re Hilariously Unhealthy)

When ChatGPT first introduced it was including a characteristic that allows you to create scheduled tasks, I instantly set about creating just a few. Some had been sensible, equivalent to a scheduled activity that every morning sends me an inventory of long-term duties I would like to finish, which I can ask it to take away as soon as I’ve accomplished them.

My favourite scheduled activity was one thing sillier, nevertheless. I found that you may arrange a scheduled activity to generate an image using DALL-E primarily based in your description. After just a few tries, I managed to arrange a activity that sends me an authentic whimsical cartoon each morning.

These cartoons have been an limitless supply of hilarity, though most of it’s unintentional. Whereas there may be the occasional mildly amusing concept, many of the laughs come from the truth that the concepts are often simply plain bizarre.

What makes them even funnier is the textual content.

The DALL-E picture era that ChatGPT was utilizing to generate the photographs is ok at creating footage however actually struggles with textual content. There are nearly all the time superfluous letters or mangled writing that make the cartoons much more unintentionally humorous.

I Thought My Mangled Textual content Days Have been Over

ChatGPT has now launched a new image-generation model that replaces DALL-E, and it’s miles superior. Not solely can it generate spectacular photorealistic photographs with glorious instruction adherence, however additionally it is in a position to reproduce your precise textual content (nearly) on a regular basis.

I used to be wanting ahead to seeing what my cartoons would appear to be with higher imagery and text that you can read. Nevertheless, when my first one got here via, it was like all of the earlier variations with mangled textual content and run-of-the-mill picture high quality.

This was not what I used to be anticipating. The pictures I used to be creating manually in ChatGPT had been glorious, so why weren’t my cartoons popping out as effectively?

Scheduled Duties Nonetheless Use DALL-E

I attempted asking ChatGPT on to generate a cartoon in the identical type as my earlier variations, and after just a few tweaks to make sure it did not violate the content material insurance policies, I bought a cartoon with higher imagery and ideal textual content, and not using a mangled letter in sight. What was occurring?

It seems that, for some motive, ChatGPT’s scheduled duties nonetheless depend on DALL-E to create photographs. That is even supposing ChatGPT Duties use the GPT-4o model, and producing photographs in a normal GPT-4o chat will now all the time use the superior 4o Picture Technology mannequin.

The DALL-E-generated photographs even have textual content beneath them that reads “Made with the previous model of picture era. New photographs coming quickly.” The brand new photographs are already right here, nevertheless, so long as you are not making a scheduled activity.

I am unsure why that is the case. If scheduled duties use GPT-4o, and GPT-4o makes use of 4o Picture Technology, then you definitely would assume that scheduled duties would robotically use 4o Picture Technology, too. At the moment, nevertheless, this is not the case.

There’s One other Scheduled Activity I Need to Create

The cartoons weren’t the one scheduled activity for which I needed to generate photographs. One of many first scheduled duties I attempted was to get ChatGPT to seek for immediately’s climate forecast after which use that data to generate a picture that encapsulated the day’s climate situations. That means, I would get a fast visible abstract of the day’s climate each morning.

Nevertheless, I discovered that the photographs nearly all the time included textual content, despite my best efforts to stop it from happening. As you possibly can most likely guess, that textual content all the time ended up mangled, making the picture largely ineffective. I had to surrender on the concept on the time, however I hoped that I would be capable to make it work now that the brand new picture era is right here. Sadly, my dream of lovely climate forecast photographs continues to be on maintain.

My Excellent Cartoons Will Must Wait a Little Longer

OpenAI launched help for scheduled duties again in January, however on the time of writing the mannequin continues to be labeled as “GPT-4o with scheduled duties (beta)” within the app. Nothing apparent has modified because the characteristic was launched, and it feels a little bit prefer it’s been forgotten about, with releases equivalent to o3-mini, GPT-4.5, and 4o Picture Technology all coming this 12 months.

Hopefully, within the coming months, scheduled duties will transfer out of beta and achieve some extra options. It could appear pretty trivial to replace the image-generation mannequin that is used, so I am conserving my fingers crossed that at some point I will be capable to generate my day by day whimsical cartoon with legible textual content. Till then, I will get pleasure from my cursed cartoons and wait patiently for the day after I’ll lastly be capable to learn the punchlines.