Talk:Midjourney

Reads like an advertisement?

I love Midjourney as much as the rest of us nerds, but this page reads a bit like an ad. I think the style should be edited to incorporate a more neural tone. Kleinhern (talk) 05:42, 1 July 2022 (UTC)

That's because it is one. Click carefully or you'll be charged by Discord (great name, that's what it sows). 77.147.79.64 (talk) 20:11, 14 October 2022 (UTC)

Yes, this is blatant advertising on Wikipedia. 2600:1702:2350:9F80:BD2D:56D0:3F5F:FA99 (talk) 19:33, 24 October 2022 (UTC)

Labels and Definitions.

A "proprietary" customization of Diffusion Notebooks amalgamated from what various contributors have been coordinating and running in Google colab/Jupyter Notebook services.

(The packaged end-product for customers)Hooked up to a Discord chatbot/server as a service after viral Twitter mirroring of Dall-E Mini.

The claim to technology would not be attributed without notable mention of RiversHaveWings CLIP+VQGAN and Somnai's addition to QOL improvements for DiscoDiffusion and JAX notebooks which openly and socially provided the conceived execution and improvements that followed.

Ironically, it is obfuscated on behalf of commerciality. 104.33.225.239 (talk) 17:45, 10 July 2022 (UTC)

Shouldn't Théâtre D'opéra Spatial be its own article?

A winning art piece that has recieved plenty of coverage, many of the sources comment on the art piece itself. I believe it should be its own page, shouldn't it? Unspectrogram (talk) 17:14, 17 September 2022 (UTC)

Sure. You've created 27 stubs, why don't you create it? I'll bet other editors will add to it. I would when I have the time. Carlstak (talk) 20:04, 17 September 2022 (UTC)

What is the source of Midjourney's training data?

Seems like a basic fact that should be in the article. Yudel (talk) 14:36, 20 September 2022 (UTC)

Agreed. There's an uncorroborated statement in the article, but that's it at the moment. Mikewyantjr (talk) 18:52, 6 December 2022 (UTC)

Who owns Midjourney?

It is important to add the owners of the company and the country they come from. 211.30.33.236 (talk) 04:42, 24 November 2022 (UTC)

Ah that would be Midjourney. Midjourney, Inc. specifically. 192.77.12.11 (talk) 07:50, 2 April 2023 (UTC)

Midjourney Training Set Citation Needed

Was reading through this and saw the citation at the end of "Because Midjourney's training set includes copyrighted artists' works, some artists have accused Midjourney of devaluing original creative work..." leads to an article that does not reference the first half of the sentence--"training set includes copyrighted artists' works." This lends to the appearance of validation of that claim without proof. I am marking the first half as needing citation unless there's a citation elsewhere that can hold up the copyright statement. Mikewyantjr (talk) 18:47, 6 December 2022 (UTC)

Photo gallery

The current photo gallery section seems like a distracting and inflammatory choice of subject, but the idea of having a section with images comparing the different releases seems reasonable.

What should go there instead? I'd lean towards something fairly standard (e.g. an avocado chair, since that became a famous "benchmark prompt" after the first DALL-E back in Jan 2021) Theseriousadult (talk) 03:09, 29 December 2022 (UTC)

Done and done --gejyspa (talk) 04:32, 29 December 2022 (UTC)

Revert. 1. Art is often controversial. See: Piss Christ. 2. The original gallery is based on a portrait of a person who has been in the public eye for five decades. 3. The photo prompt is a work of the US government, therefore public and no dispute of copyright or trademark. 4. The disparity between AI output of v1 and v4 is significantly more evident in the original gallery than in the avocado chair gallery. 5. The opinion of the topic's developers are a conflict of interest. --Chicago god (talk) 04:37, 29 December 2022 (UTC)

I concede point 5.

As to point 1: Yes, but the avowed intent of the artist of Piss Christ was to be controversial. That is not the intent of this gallery, which is to show the differences of the different versions. To deliberately choose a picture of a controversial figure clouds that intent.

Point 2 has, as well, no bearing on that same issue. The photo could be of anything in order to highlight the differences. So why does a picture of President Trump have any more or less claim as the exemplar subject? And further, the pictures of Trump are by admission of their submitter were NOT all produced with the same prompt. So the avocado chairs show off the difference better

Point 3: The creator of the pictures of the avocado chair has declared them to be CC 4 BY-SA, and I believe has submitted such an email to the permissions commons of Wikimedia

Point 4: I concede the point that the differences are more evident in the Trump pictures than the chair. But the quality of difference is also are dependent on what prompt was used. Again, since a choice was made, on what basis do we say one is more "right" than another? --gejyspa (talk) 10:51, 29 December 2022 (UTC)

To further clarify point 1, the intent of the gallery was to show the differences of the versions using a photo prompt, and then to finally show the difference in version 4 with and without a photo prompt (but using the same text prompt). Also, aside, the Piss Christ article here doesn't seem to share your opinion that it was intended to be controversial.

Point 2, of the five images in the gallery the first four used the same photo prompt and text prompt. The only change to the text prompt was to force Midjjourney to use a specific version of the AI engine (e.g "--v 1"). The fifth image didn't use an image prompt, but used the same text prompt -- here the intent was to show how much weight the version 4 engine places on the image (very heavily).

We already have a meeting of the minds regarding points 3 and 5.

Point 4, while any art is going to be subjective, I think the visual disparity in the difference between v1 and v4 is much more apparent in the Trump gallery, and is even more impressive when one considers that all of the AI engine versions were released over the course of just one year.

The ability of the engines to translate text to photorealistic artwork is better represented by the portraiture than by a still-life, IMHO. Chicago god (talk) 11:25, 29 December 2022 (UTC)

While subjects certainly do not get to dictate the content of articles, I don't think going out of our way to piss off subjects is a particularly good way to maintain encyclopedic standards. Common sense should be used. jp×g 16:38, 29 December 2022 (UTC)

Piss Christ would actually be a great choice here, as would "President Donald Trump". Putting him in Bozo the Clown makeup doesn't seem in line with neutrality. Theseriousadult (talk) 05:10, 29 December 2022 (UTC)

I tried Piss Christ, but the word "piss" is banned as a prompt in Midjourney and I don't want to try to circumvent their filter. Chicago god (talk) 11:00, 29 December 2022 (UTC)

I suggested "an avocado chair" because it's a standard benchmark prompt in the field. Another standard choice that might work better is "an astronaut riding on a horse"? Theseriousadult (talk) 05:11, 29 December 2022 (UTC)

My concern is that precisely because those two prompts are better known, the AI engines would've potentially been better trained to create them. The Trump portrait image prompt with Bozo text prompt would've required more "effort" (for lack of vocabulary) on the part of the engines. A quick Google search found 24M avocado chair and 3M horse-riding astronaut results, whereas Trump Bozo Makeup is around 400K results. Chicago god (talk) 11:33, 29 December 2022 (UTC)

I think that if we put our brains together, it should be possible to come up with some illustrations that showcase the development of the engine (perhaps including a person) without slathering the page in goofy politics stuff. The fact that WP:NOTCENSORED does not mean, for example, that Photography needs to be illustrated exclusively with photos of Joe Biden, or Joe Biden's genitals, or whatever. It may well be the case that the avocado armchair is boring, in which case, I dunno, we could still come up with something a little less silly. Benjamin Franklin flying a drone? A polar bear reading a book about apples? Who knows. jp×g 11:47, 29 December 2022 (UTC)

I agree with avoiding the "goofy politics stuff." The current group of avocado chair images is a vast improvement over the previous series of images which seemed designed to express a wikipedia editor's non-neutral personal point of view about a living person. Elspea756 (talk) 16:04, 29 December 2022 (UTC)

Censorship

I am surprised that Midjourney censors quite a lot of prompts (beside pornography or obviously illegal stuff). Following examples:

Regarding Communist China: "Chinese president", "Xi"

Regarding Islamists: "Prophet Muhammad" A11w1ss3nd (talk) 10:21, 21 March 2023 (UTC)

The blocking of images of Xi is now covered in the article. Blocking of images of Muhammad (when the prompt makes it clear they are intended as such) would presumably be for a similar reason - that is, allowing such images might get Midjourney banned in some countries. But we would need a reliable source that such blocking is done before we could mention it in the article. --Zundark (talk) 18:16, 9 April 2023 (UTC)

Image License

The mechanical dove image on this page is licensed as CC-BY-SA, yet the teddy bear image on the page for DALL-E is listed as being in the public domain, "as the work of a computer algorithm or artificial intelligence, it has no human author in whom copyright is vested." The same is true for Stable Diffusion's image. Is there a reason for this difference? 64.213.134.4 (talk) 17:07, 3 May 2023 (UTC)

Corrected Trade (talk) 19:46, 21 May 2023 (UTC)

Can we please lock the article?

Getting tired of people constantly adding their "art" to the gallery solely to keep it from being deleted on Commons. Or at the very least make a rule that additions or images to the article have to be discussed on the tal page first. Trade (talk) 19:57, 21 May 2023 (UTC)

It seems redundant to have one per WP:GALLERY, when the article can have relevant images in the text. I've added in images for Théâtre d'Opéra Spatial and Alice and Sparkle.

If we're avoiding the fake Macron/Trump/pope viral images because we can't know for sure that they weren't also edited by a human, can we generate a free version as an example? Belbury (talk) 10:44, 14 June 2023 (UTC)

Actually the linked Trump image is already on Commons at File:Trump’s arrest (2).jpg as public domain, cited to a Twitter thread where a user is rapidly generating a bunch of them, so I guess it's taken as straight Midjourney output.

I'll add it to the article, and I think that in combination with the two examples above removes the need for a gallery section. Belbury (talk) 10:51, 14 June 2023 (UTC)

Yes, thanks to both of you, the current version with notable examples is a great improvement over the previous gallery of non-notable user-generated image spam. I will note that the "Alice and Sparkle" image has what looks like clear human-created graphic design in the title, so it is likely not public domain. I'll try to replace it with an interior page, cropped to just the image, from https://time.com/6240569/ai-childrens-book-alice-and-sparkle-artists-unhappy/ when I get a chance. There is a good example captioned "Artists singled out this page from Alice and Sparkle as showing the limits of the AI-powered technology. The illustration has several apparent flaws, including the character appearing to have claws." Elspea756 (talk) 13:48, 14 June 2023 (UTC)

I've just removed the previous "Alice and Sparkle" image that had clear human-created graphic design, so it is not public domain, and I replaced it with another image with no human-created graphic design from the same book, which also has criticism commentary from TIME about that particular image in the caption. Elspea756 (talk) 13:23, 19 August 2023 (UTC)

The graphic design for the cover image makes no difference to the copyright status, as it doesn't come close to the required threshold of originality (see commons:COM:TOO USA). In any case, decisions about copyright status of files on Commons should be made on Commons. (There is a deletion discussion for the image in question on Commons right now. I don't know which way the decision will go, but if it gets deleted then so will all the Alice and Sparkle images.) --Zundark (talk) 20:04, 20 August 2023 (UTC)

Regardless of copyright status, this image is a big improvement over the previous image for use in this Midjourney article since it shows just an image created by Midjourney (rather than an image that also includes human-created graphic design, that might mislead readers to believe the text was written by Midjourney) and it more importantly provides us with the ability to illustrate commentary and criticism of the "limits of the AI-powered technology" as discussed in Time. This image helps us write an encyclopedia article, rather than just decorate it. Elspea756 (talk) 21:12, 20 August 2023 (UTC)

Almost all images that were provided are artificially generated. They didn't had Time's article on them though. AXONOV (talk) ⚑ 06:01, 21 August 2023 (UTC)

Hi thanks for your work @Belbury!

I think the Théâtre d'Opéra Spatial and Alice and Sparkle. images are great illustrations for the article.

However, regarding the third AI image: the "photograph" depicting Donald Trump being arrested by "policemen". I acknowledge that there is a caption below describing the image as AI artwork, and I appreciate your efforts to use it for illustrative purposes.

However, I have removed this image, as per Wikipedia policy on information on living persons in biographies and other articles (WP:BLP, WP:BLPGOSSIP etc) due to the potential for possible misinterpretation that this is an actual photograph (especially if reproduced elsewhere), and potential for defamation of a living person.

The article text already describes this image well enough, and a link is provided for readers who wish to see it. WP content is read, quoted and reproduced by search engines, children, non-English speakers and sites with computer-generated content, all of whom cannot be guaranteed to understand this is AI artwork, not a real photograph. Furthermore, although the article text mentioned that the event is fictional, the caption did not make this known.

There is also potential that misquoting or republishing of the image (e.g. if it is re-used elsewhere and attributed "Source: Wikipedia", as is common) will undermine the trustworthiness of Wikipedia, which we wish to avoid.

But thanks for your work, and I appreciate it. Fh1 (talk) 02:44, 10 September 2023 (UTC)

solely to keep it from being deleted on Commons
Has little to do with reality. I've added the gallery to display a spectrum of capabilities of Midjourney and have zero association with authors of the images (unless of course they were generated by them personally). I oppose idea of adding any viral images, political or not, they have zero relevance to this article. The motivation driving addition of WP:GALLERY was mostly lack of space in the article. It's also better to represent the subject by images rather than by text. The images need to be agreed upon of course. I think we need to remove everything and WP:RFC a broader community. AXONOV (talk) ⚑ 14:40, 16 August 2023 (UTC)