DALL-E images downloading as WebP

So, you wasted 2 GPT-4 requests to get your image, instead of one? Nice. Are you sure that what it gave you is actually PNG? Open it with Notepad, see what are the first two words.

1 Like

Don’t forget - you can always stop paying for DALL-E and start paying for Midjourney! And that’s it!

2 Likes

Chapter 3 in my fictional ChatGPT Plus Creator’s Guide is the employment of iteration abilities by prompting.

Free PNG for you to enjoy

1 Like

That, actually “worked”, thank you. 2 concerns, however:

  1. How long will it take for OpenAI to close the loophole, and
  2. The quality. Despite the instruction, it wasn’t 24-bit PNG, it was 8-bit, far from lossless. I tried it a few times, it’s always 8-bit PNG. I assume that it generated LQ WebP, and then lost even more in conversion, but conversion to PNG was quick and dirty, by assuming that 8 bits were enough.

Attached are 3 images. One from Dall-E, another from Stable Diffusion (which, ironically, generated much quicker), and last one is from Dall-E API
. Subjects aside, compare the rendering quality. The file size was 3 times different too. You may notice that both Dall-E images have similar traits, such as grainy sky and lack of fine details. It’s hardly an artistic touch, but rather a deliberate downscale and compression.



1 Like

You can look at the file size. The one I downloaded was 1.7MB, consistent with lossless compression of 3MB, and similar to saving the WebP as PNG from Photoshop.

PNG compression on 1k x 1k x 8 bit would be under 1MB.

Not satisfied, I let the AI go at it with multiple steps of analysis before reporting on what it had in its mount point with another creation.

I’ll begin with generating the new image now.

The new image has been successfully created. Now, I’ll proceed with the analysis of this image to determine its bit depth and color variety, ensuring it meets the criteria for a lossless, true-color image. Let’s start by converting the image to PNG format and then performing the analysis.

The analysis of the newly converted PNG image provides the following insights:

  • The image is not palette-based, indicating it does not use a limited palette of colors.
  • The color mode is RGB, which typically corresponds to a 24-bit color depth (8 bits per channel for Red, Green, and Blue).
  • While the exact bit depth attribute is not directly provided, the RGB mode strongly suggests it is 24-bit.
  • The image contains 162,265 unique colors, which is indicative of a high color variety, consistent with a 24-bit depth image.

Based on these findings, the image appears to be in a lossless, true-color format. To ensure this quality in future conversions, you can use the following guidelines:

  1. Use the Pillow (PIL) library in Python for image handling.

Verifying the AI assertions of color count and depth after getting annotation citation download links:

Although it uses another AI generation, the image computation in Python is almost free.

The ultimate goal may be to reduce the size of perpetual image storage in ChatGPT chat backend database, and the size of the web download location that images are temporarily published to.

The Python sandbox storage expires, so the persistent space concerns are still addressed.

Just ask GPT to convert to PNG and it creates a download link. Not big deal.

2 Likes

If I use the aspect ratio it recommends, it saves as webp and I can’t use it at all, but if I use my own ratio I’m able to see it. Right now I’ve only tested it with 99:128 and 2:3, but it seems to do the trick.
The quality is another issue entirely, oof.

This is an utterly pathetic move from OpenAI. Unsubscribed, gonna use alternative services, fortunately there are plenty already.

Best workaround I’ve found is to use Adobe Photoshop to convert from WEBP to PNG.

That’s not a workaround. You are literally just changing the filename extention

OpenAI should fix this (as well as the response issues which has me using Gemini and Poe more often than not). Having to convert the image is a dumb workaround, but that is what you are left with.
I started to write a long comment here, but my ADHD brain ended up forcing me to create an online webp to png converter. You’re welcome.

I use a free (so far) online io batch conversion program that works for me called towebp.

I also use other paid software for resizing and others for improving DPI and making enhancements, which I used when DALL-E was creating PNG images.

Based onthe number of replies, I think OpenAi should consider going back to png. As paid users I think having to do an extra step to convert images is too time consuming. Please consider changing this feature back.

1 Like

Related… I only have this problem in Windows. The iOS/iPad version of ChatGPT downloads image files as expected, so that’s where I do most of my work with DALLE.

the extension Save Image As Type doesn’t work on openai productions, it seems.
for whatever reason, they want paying customers to just see it and not use it.
Guess I’ll try Microsoft’s attempt.

$20 coverts lots of things so that’s why it’s better to pay here for beginners.

Adobe seems to agree with us on this one. I did what they said about changing the file extension name to .png but Adobe’s calling out my bluff xD
image

1 Like

its easy to take DALL-E and ChatGPT for granted. We’re on the bleeding edge people.

This converts perfectly

When you download images directly from the interface, they are indeed in the WebP format, which is known for its efficiency and quality at smaller file sizes. To get these images in PNG or JPEG format, you would typically use a conversion tool or software. Here’s a simple workaround using online conversion tools:

  1. Download the Image: Click on the image you want to download; it will save in WebP format.
  2. Use an Online Converter:
  • Search for “WebP to PNG online converter” or “WebP to JPEG online converter” in your web browser.
  • Select a tool from the search results. Many websites offer this service for free and without the need to sign up or install anything.
  • Upload your downloaded WebP image to the converter.
  • Choose the output format (PNG or JPEG) as required.
  • Convert the file and then download the converted image to your device.

This method doesn’t require any special software, and it’s quite straightforward. If you prefer using software, image editors like GIMP or Photoshop can also convert WebP files to other formats.

I am guessing that the reason they use WebP instead of one like PNG is due to WebP images being smaller in size than PNGs, meaning they can host more of them on their servers.

1 Like

I installed a “save as PNG” Chrome extension that does not work.