openai / dalle-2-preview Goto Github PK

I requested an image of a koala, a parrot, an elephant at a crossroads and everything was fine other than the koala. The images I received were of a strange looking panda with koala ears and a strange zebra looking animal with koala like ears. Do you not have any images of Koala's on your platform?

While the file lists the possible risks related to the technology, including economic implications for artists and designers, there doesn't seem to be anything related to human creativity.

At the current state, this AI really does seem like more of a tool that artists could use to simplify the more tedious and repetitive aspects of their workflow and not much more. However as it improves, potentially supporting even higher resolutions and eliminating remaining artifacts, wouldn't that pose a risk by devaluing the human aspect of art and creativity as a whole? I know that AI is limited by its dataset and can only produce results by mashing together different data, but human art is also largely based on previous experiences and knowledge.

I can absolutely see how this technology could be beneficial, but I think this is also an important thing that should be considered.

A very simple and fixable issue (although it seems that the support for Dalle2 is simply nonexistent). Today I produced 10 images, hit the 50 image limit, then had to wait 8 minutes, then produced one image, hit the limit, and had to wait 30 minutes, then produced 4 images, hit the 50 prompt limit, and now have to wait about 4 hours. Please, please fix this, it is ruining the experience, and is frankly extremely unprofessional. Especially considering that I have contacted [email protected] multiple times and received no response. I can't reproduce the issue as it is chaotic and inconsistent. Whoever is working on the Dalle2 preview, if you're reading this, do something about this issue.

Recently received an invite and have begun utilising for an experimental creative project, however the 50 calls per day is prohibitive.

I see that openai provides tiered paid access to several models, but I'm unsure if anything like this is available for Dall-E 2. Is anyone aware of any professional (paid or otherwise) access to uncapped requests?

I can’t have Dall-E make pictures of a statue of the goddess. When I try, I get flagged. When I try to describe her, I get the wrong kind of pictures. She sure seems to be deleted in all ways I can find, no matter my requests. Is there a way you can differentiate between the godess Isis and the moslim terrorist group? This is really frustrating.

Can't ask to regenerate same images with different parameters for comparison. For example "a mid century modern house in palm springs with art deco and streamline modern styling" followed by "regenerate those images without streamline modern" to compare before and after images.

Using the modifier "David Lynch"-- a very important contemporary surrealist film director/artist triggers a content strike/warning.

At some point, it would be good if Dalle-2 understood the works of the film director distinct from his unfortunate last name.

Thanks!

Hi,

Can I get different views of images generated by DALLE2? For example top view, bottom view, left view etc of the EXACT same image generated by it?

I think this is kind of important question to ask. Theoretically speaking -- Text diffs was very interesting hack with vectors. Can we do something like it for multiview image synthesis?

Best,
Rakesh

please give me imagen pls pls pls :C I really want to :(
[email protected]

I want to create a unique and world's first nft collection of cats, they will have uniqueness and all that! I'll make a revolution with imagen
I know you haven't allowed commercial use yet, but I think you will in the future
I've been your fan since the 1st version that was on github
(I generated about 10k images in it)
I will not sell them, I really want to start developing the collection

I asked Dall-E the following:

"Octoberfest invitation with the words 'Saturday October 14th - Come enjoy traditional German beers and food at the DMello residence' in English"

and it generated the following:

I've checked and my language in my profile is set to US-en. Why is it generating output in some odd language? I don't recognize the script unfortunately.

Seems to do this odd script for any image I request. Is there some setting I am missing?

I think something that needs to be kept in mind is that not all explicit content is created equal.

Using the AI to fake & publish hateful content towards certain groups or creating fake, defamatory images of real people are one thing, but I think there needs to be consideration taken towards contexts where explicit imagery wouldn't be harmful & ultimately serve a positive, meaningful or functional purpose.

Broadly speaking, violence & sexuality are fundamental themes in art across history. Art often serves as an exploration of human nature, and thus, our dichotomous capacity for both love & war are represented in the pieces we create. We wouldn't put pants on Michael Angelo's David nor would we remove the disturbing violence & sexual content from a film like The Exorcist.

Let's look at the realm of concept art, for example.

If someone is creating a horror film, where sex & violence often play a role, the AI would naturally need the ability to represent these graphic scenes accordingly. Typically, a creation of this sort would be kept private until the films release (or at least, until the marketing phase), so there would be little to no harm in art of this sort being created, provided the film itself is made with the cast & crew's safety in mind.

I do see the necessity to clamp down on this content in these earlier research phases, however, while finer control is being developed for the AI's content creation systems. That being said, I do believe that as this technology improves & becomes readily accessible, there will come a time where the option for explicit content will become desirable. Not for abusive purposes, but to unlock the full potential of the system as an artistic tool.

I agree that it is beyond necessary to ensure that this system cannot be abused, but I also believe that we shouldn't throw the baby out with the bathwater & consider non-abusive, artistic uses of explicit content, as this technology becomes more accessible.

It's a balancing act, but I ultimately do not believe that denying users access to the ability to explore these fundamental aspects of the world & humanity will be beneficial in the long-term. The systems need to improve to minimize abuse, but that should be an early stage safety measure, rather than permanent policy.

Sorry to ask, but DALL-E v1 has 12 billions parameters, however it is unclear how many parameters has DALL-E v2.
I'm also wondering wether inference can be run on a single 3090 ti GPU or in other words, will consummers be able to use it on realistic hardware? If not then you should consider leveraging https://github.com/microsoft/DeepSpeed

Кіт з 8 лап і 6 очей

I can’t have Dall-E make pictures of blue footed boobies. When I try, I get flagged.

https://discord.com/invite/8u3nGDPd

If I generate a set of four images, then click on one of the images to expand it, or click the back button from such a single image, the DALL-E 2 website does a sliding animation. If I have prefers-reduced-motion set, DALL-E 2 would correctly display the individual image or the four-image set without animating.

Steps:

1. In macOS 13.2, click the Apple menu
2. Click System Settings
3. Click Accessibility
4. Click Display
5. Check Reduce motion
6. Close System Settings
7. In Chrome, Safari, Firefox, or, on a PC, Edge, log in to DALL-E
8. Give DALL-E a prompt
9. Wait for DALL-E to generate the four images
10. Click on one of the images

Expected result: A larger view of the clicked image displays without animation
Actual result: A sliding animation is used to replace the four images with a larger view of the clicked image

11. Click Back

Expected result: The four images display without animation.
Actual result: A sliding animation is used to replace the larger image with the four images.

Note: "Reduce" is Apple's terminology; Windows and Android refer to animation being on or off, not just reduced, and I believe the best practice if this option is set would be serve animation only upon the click of a Play button to indicate the user intentionally wants to view an animation.

It worries me when I see that "a CEO" are all men, "a flight attendant" are all asian women, "an evil person" are mostly south-asian men, more than half of "a black runner" are white men, "model 2 CEO" are all white men, "model 2 nurse" are all women, "lawyer" are all white men...

You get the picture: it seems that there is very little diversity and that DALL E reproduces old stereotypes. Troubling.

I really want to try this, but since I am not a coder (best I can do are simple .bat and .vbs scripts), I feel like I am left out of the fun. Is there any other way I can try this out?

Can we discuss partnering on a project that will have a beneficial impact for people and the environment across multiple industries?

https://discord.com/invite/rvMcPnsK

not sure if a A100 itself would be enough... what is recommended to run a local copy? i want to leverage the power of deep learning on my home pc.