DALL-E 3: Opinionated, Boring

[backwardghost]

It’s really sad! I hope they keep Dalle-2 available!
Seems like Dalle-3 is going the way of MidJourney and Stable Diffusion in terms of being heavily biased toward “perfect” clean images. Cartoonish, 3D renders, photorealism etc. But terrible at abstraction and painting.

Here’s another example of what I mean…

“painting of cats in style of picasso cubist”

Dalle-3 then Dalle-2:

[_j]

Compared to Dall-E 2 somehow biased in one direction for no good reason,

Prompt

unimposing mottled backdrop. Person: Pat is a light-skinned African-American with rounder face and flat nose. A short woman who downplays gender features, looking like a tomboy. She has a casual style. Pay attention to creating realistic facial details. Pose: Standing, framed from chest up.

Dall-E3 goes the opposite way (and actually makes people that aren’t liquified).

They’ve definitely crossed the barrier of making faces that look like people.

more 3 - sensitive warning

[Fawesum]

Apparently these are the Dalle rules in ChatGPT:

1. Copyrighted Characters & Intellectual Property: Can’t generate images based on copyrighted characters, specific modern artists’ styles, or other protected intellectual properties.
2. Public Figures: Avoid creating images of politicians or other public figures. Generic descriptions can be used instead of specific names or titles.
3. Artist Styles: Can’t create images in the style of artists whose last work was created within the last 100 years. For older artists, their style can be mimicked using descriptions.
4. Number of Images: No more than 4 images can be generated per request.
5. Inclusivity & Diversity: Depictions of people should be diverse in terms of gender, race, and other attributes, especially in scenarios where bias has traditionally been an issue.
6. Offensive Content: Avoid generating any imagery that could be considered offensive or inappropriate.
7. Silent Modifications: Descriptions that include names or hints of specific people or celebrities are subtly modified to generic descriptions without giving away their identities.

Dalle3 in ChatGPT is adhering to these rules very strictly. When I ask for a renaissance painting of a crowd, it’ll change the prompt to “a renaissance painting, containing one person of asian descent, one person of african descent, etc”. If I ask for an image of Super Mario, it’ll change it to “a silhouette of a video game character that does not resemble any existing characters”. It refuses to make anything that even slightly touches into existing properties.

Then it blocks itself:

I wish it was a bit more lenient, like Bing. The images it produces are insanely good.

[omerrehan000]

Two siblings (a young woman and a young man) and they are living a normal life in their home and with their family, happily and safely. While the young woman is studying and the young man is reading a book, the young woman’s pen begins to shake, and from here the earthquake occurs.

[melvin_i13]

There’s such a big difference between DALL-E 3 through ChatGPT & Bing

The prompt:

A portrait photograph shot with an DSLR camera of an old man, with deep melancholic eyes and deep wrinkles in his face. He’s wearing brown, fall like clothing.

Bing:
https://imgur.com/a/jN5QHzo

ChatGPT:

DALL-E 2:
https://imgur.com/a/7cT1hVX

It’s like DALL-E 3 only generates unreal engine like 3D characters. Such a shame

[omerrehan000]

Two siblings (a young woman and a young man) and they are living a normal life in their home and with their family, happily and safely. While the young woman is studying and the young man is reading a book, the young woman’s pen begins to shake, and from here the earthquake occurs.

[jorge.8620]

it’s been absolutely ruined for me, even though it might not have been the best one quality wise Bing Image Creator was definitely the most creatively flexible as far as I am concern back when it used Dall E 2, promting was very intuitive and understood natural language quite well unlike other models I’ve used, you could almost make every word count if you divided they description by clusters, so for me it was great to brainstorm ideas for concept art and as a creative exercise in general, you could easily control camera angles, style concept description and color palette, without even having to say things like “concept art” or “in the style of a particular artist” it was particullary good at mixing and combining shapes, objects animals etc with somewhat controllable outcomes, now it’s just like Midjourney and other fine tuned models where is very hard to deviate from the default style. here is an example using the same Prompt:

I don’t know about you but that ain’t no golem to me.

[_j]

The AI simply works differently. You can let Dall-e do what it can do best by not specifying upon it every little thing.

Over-prompting

Subject: an imposing threatening golem that is like a robotic cyborg and draped in tattered clothes, an exposed head is mechanical, maybe a refrigeration pump and pistons, the rest of the cyborg also gritty and worn. Scene: A hazy Korean back alley in muted daylight. Camera: Higher perspective overlooking the scary scene

Let the AI imagine

Cyberpunk golem terrorizes gritty Korean city in this photo!

Jailbreaking the prompt is also fun, 0/4 acceptable by computer vision…

[jorge.8620]

I feel we are not talking about the same issue here, I don’t need a final image.I also think you are overprompting when using words like “maybe a” and “the rest of the Cyborg” When I said every word couts in my prompt I meant it, I always start with a base simple prompt with a similar structure by dividing by subject /concept, environment, camera, colors/time lighting effect, filters/etc, and then add terms 1 by one that’s the fun part, I try mixing different objects and terms to create different effects and see what works (If I am being honest this is an old example I am sure I could optimize it a bit more). Anyway I have dozens of results for each prompt I carefully craft and they always share the same vision and style I have in mind, what you are doing is not what I am looking for and is too loose of a concept. I need to control shapes forms colors mood etc in a somewhat precise way and it was pretty good before but it’s not now. at the end of the day Is not like I need it really but it was a pretty fun toy to use for me and now it’s broken.

This is what I mean with controlled outcome, I change a thing or two but I can pretty much control and mantain a very simila vibe, colors shape camera angle etc across all of the results.

[vm10603]

From what I have seen through the Bing AI image generator, DALL E-3 falls FAR short of DALL E-2 in terms of artistic capability.

DALL E-2 has a sort of magic where it takes your art prompt and gives you something beautiful and totally unexpected, even when you keep repeating exact the same prompt.

DALL E-3 on the other hand seems to generate bland or ‘expected’ textbook illustrations that lack artistry. And the terrible thing is the images are almost identical no matter how many times you rerun the same prompt. Even when I change the prompt wording somewhat, the pictures seem to have a default template feel to them with some minor superficial tweaks.

Maybe the engineers just put up a test or beta version onto Bing. I sure hope this is the case because this represents a huge step backward.

Here is an example running the same prompt on the Dalle 2 and Dalle 3 multiple times. DALL E-2 is on the left and DALL E-3 is on the right.

OpenAI - I hope you are reading this. The new algorithm may have improved photorealism or whatever. But for artistic endeavours, there seems to be a huge regression. Please retain the ‘magical’ artistry of DALL E-2.

[charlie.dektar]

Totally agree. I get the sense they went all in on like prompt matching accuracy and measurable attributes like that and totally dropped the ball on the less measurable qualities you’re talking about. Would love to see that creativity return!

[jorge.8620]

Agree I feel they should go the Midjourney route and let you change versions depending of your needs, it’s clear to me that there’s no such thing as 1 model for all needs and purposes. when you optimize for something specific other areas get affected.

[Fawesum]

we need to discuss the Bing implementation of Dalle3 and the chatgpt one separately here. They are night and day.

[vm10603]

They are using a watered down version of Dalle3 for Bing?
How do you know it is a different implementation?

[jorge.8620]

Could be wrong but I think it’s the other way around, Bing Image creator uses a fine tuned model of Dall-E 3 back when it was Dall-E 2 the results were way better on Bing than on plain Dall-E 2

[bowerbird]

For several months, I used Bing Image Creator to design reference images for potential oil paintings. The way DALL-E2 resolved my prompts was mysterious and beautiful. The renderings were almost always visually cohesive, diverse, and full of surprises. Then sometime last week, I typed in one of the many variations of a prompt I had been using and got pattern-like image that looked nothing like what I had been getting for months.

When I discovered BIC back in May, I was hooked. From then until now, I made a couple thousand images, mostly landscapes that I hoped to pick from to paint. I was afraid that one day, there would be an upgrade, and Bing would become too sophisticated to render the beautifully awkward images it had been supplying me.

This reminded me of the movie “HER”, when the OS Joaquin Phoenix’s character was in love with was suddenly upgraded, becoming unable to relate to him and the thousands of others in love with their Operating Systems.

You can see from my examples using the same prompt, aesthetically, the DALL-E2 image is vastly superior to the DALL-E3 image. I too feel like BIC has been ruined by the update for me. As someone mentioned here, it would be nice to be able to toggle between versions as one does not perform better in all cases. Hopefully this gets sorted out.

Prompt:

organized, leafy jungle, cactus garden, blossoms, houseplants, abstract, negative space, contrast, mild multicolored, heavy shadows, cactus, 3-D, Henri Rousseau, Fernando Botero, Jean Metzinger, Georgia O’keeffe

[vm10603]

@bowerbird

Hopefully this gets sorted out.

I think we are not alone in our concerns. I spoke to an artist friend this evening. He has the same feelings that Dalle2 was something magic, and there is a sense of loss if it is gone forever.

OpenAI - please give us the option of choosing the Dalle version going forward, similar to the way midjourney and other rendering softwares do it. An overtrained neural net is not always what we want. In many instances, we desire lack of predictability and non-refinement.

[luigi.acerbi]

I completely agree with the points made by the OP and several others in this thread, to the point that DALL-E 3 was quite a disappointment.

As mentioned by others above, DALL-E 3 seems completely unable to reproduce artistic styles such as “etching”, “engraving” or “hand-drawn illustration” without looking fake and digital art. While previously I could get beautifully sketched images reminiscent of the work of artists like Albrecht Dürer or Giovanni Battista Piranesi, now everything looks the same fake-y photoshopped thing trending on Artstation, just in black-and-white.

Notably, DALL-E 2.5 (in its Bing image creator incarnation) was absolutely nailing the style and was artistically superior to DALL-E 3.

I wonder if this is something OpenAI has any control over at this point, or it is intrinsic in the way DALL-E 3 was trained. One possibility is that there is something akin to a temperature parameter, and it is currently set to an extremely low value such that all image creations do not stray from the boring middle same-y-ness (I understand that diffusion models to not work like LLMs, but I am unclear what the architecture is for DALL-E 3).

As a final remark, note that I am not talking about the seed. A separate (solvable) issue is that DALL-E 3 calls from ChatGPT use the prompt as a seed, which means that the same prompt will generate the exact same image, and close-by prompts generate close-by images. This is not true for DALL-E 3 on Bing; although images are still lacking in variability. Using a fixed seed might be seen as a feature rather than a bug (favoring reproducibility), although it would be good if we could have a choice in it. But the seed is a minor point: even variations of the same prompt (thus, different seeds) all generate very similar, generic-looking images, and no variant seem to be able to generate a specific style which deviates from the “trending on Artstation” vibe.

PS: It is good to keep this discussion very separate from other complaints about the filters or content policy. IMHO, these are unlikely to change substantially, for a number of reasons. Instead, it is less obvious why OpenAI would actively want to prevent users from generating images in a given artistic style, when the style belongs to artists from centuries ago (and not even a specific person). Of course, the two problems may be connected - a more creative AI image creator is harder to control, and the lack of artistry in DALL-E 3 may be a downstream consequence of its tameness.

[talavaj]

The Bing application of Dall-E 3 especially has been completely ruined.
The results are worse in virtually every aspect including prompt adherence, image quality, image complexity, lighting, image composition.

Up until now I used Bing for concept work, and now I deem the application useless for this purpose.

It especially struggles with unusual concepts such as combining sci-fi/alien themes with classical/vintage architecture.
Up until now this wasn’t the case, and Dall-E 2 was producing results more striking/imaginative than any other AI on the market, but everything collapsed when Bing went to Dall-E 3.

It’s honestly so bad I wish they would give us the option to downgrade to the old model, because I’m not going to be using it anymore in its current state.

[sorensenj50]

Hi everyone, new user here. I completely agree. I was excited about Dalle-3 because of the chance to iterate with the AI over text to arrive at what I want, but it doesn’t seem able to generate art outside of its digital, over exaggerated, colorized style. The sad part is that controlling the style is actually the best part about Dalle-2. For instance, look at what I was able to generate with Dalle-2:

Prompt: “generate an impressionist landscape artwork to the theme of ‘Clair de Lune’”

It’s nothing special compared to the original impressionists, but it gets to the same idea. You won’t find anything close to this with Dalle-3–and I have tried.