X got a new AI image generator called Aurora | Mashable.
X got a new AI image generator called Aurora
Aurora is here and you can try it for free. Credit: Jaque Silva/NurPhoto via Getty Images |
X is going in even more on AI.
A new AI image generator called Aurora started rolling out for X users this weekend, per TechCrunch. Aurora is part of the Grok AI tab along the bottom of the X app, and can generate realistic images from written prompts. The weirdest part is that Aurora appears to have no guardrails around creating photo-realistic images of real people. I was able to create an image of the upcoming president-to-be covered in cheese, for example.
SEE ALSO:Apple's iOS 18.2 arrives soon: Here's 4 new features to be excited about
It won't do anything nude, but it'll do more than the average AI image generator because most mainstream ones will prevent users from including copyrighted works or real people in their prompts. X's Grok AI has had image creation capabilities for a while now, but Aurora appears to be a new creation engine from the previous one. It's also not clear where exactly it came from, who exactly made it, and what exactly it's been trained on. It's been reported in the past that the Grok chatbot trains off of user data, so it's possible Aurora does, too.
Try it if you want, or don't. The world is your oyster.
Topics Artificial Intelligence X/Twitter
No example of how photorealistic it is?
ReplyDeletesome examples
Deletehttps://x.com/SawyerMerritt/status/1865443191295549470
https://x.com/Yasin__Shafiei/status/1865346588425478422
https://x.com/MaryTilesTexas/status/1865494355122835721
WOW 🔥🔥
DeleteAurora make an image of mashables Facebook staff.
ReplyDeletehttps://media1.tenor.co/m/t612vbmCp9UAAAAd/stare-weird.gif?
Az az xAI Aurora kepgenerator?
ReplyDeleteI wonder how many people grok what it's like to be a "Stranger in a Strange Land"?
ReplyDeletewonder if these cats know anything about Grokking?
DeleteJust tried the same prompts for images using both ChatGPT and Grok 2 + Aurora. Grok has definitely been better.
ReplyDeleteWhen I click on the Aurora button it jumps back up to the first button automatically. Maybe they're updating it right now!
DeleteOh, I just understood. It's working in the web version, but in the app, it's the same as you mentioned—switching to the previous model.
DeleteIt's sticking to the new one for me; however, if I press F5, it reverts back to the regular Grok.
Deleteclear! Flux is a generation or two above Dalle3
DeleteThe photo realism is wild
ReplyDeletehttps://imgur.com/a/amwcAIN
https://imgur.com/a/8UwOx2r
Deletedamn. that’s crazy.
DeleteThis comment has been removed by a blog administrator.
DeleteThis image is not good.
DeleteThis comment has been removed by a blog administrator.
ReplyDeleteIf you have access, please ask this model to generate a peeled banana. I haven't been able to do that yet.
ReplyDeletehttps://imgur.com/a/9J8V72X
DeleteHere it is... what do you think?
thats the best looking peeled banana ive ever seen
DeleteSucceeded. Can't say it has been as natural as I'd like it to be, but it looks pretty good.
DeleteI fear labs like Midjourney and Runway will be out of business soon.
ReplyDeleteI bet it's just the newer version of Flux. Grok doesn't have its own generator, it's always been using Flux and they recently released upgraded versions of that
ReplyDeleteIt is in fact not. @Miaouwo on X works for xAI and literally just now told me it is in fact their own model.
DeleteLink for what it’s worth: https://x.com/miaouwo/status/1865485477622509829?s=46
Moved to Bluesky from Meta and Twitter. Its a breath of fresh air.
ReplyDeleteBut can it do a plain white square?
ReplyDeleteas someone who is wary of elon musk and all things connected to him due to him being a net negative for humanity itself and democracies worldwide while simultaneously propping up dictators everywhere. I personally have no interest in grok now or ever and honestly hope it fails
ReplyDeletethank you for letting the world know
DeleteThank you for thanking them.
DeleteFinally, AI can make a photorealistic image of Garfield holding a gun. I take it all back, this is a useful tool now.
ReplyDeleteGood, because the insane censorship is so over the top that the other image generator models all look like stupid cartoons.
ReplyDeleteI noticed the hurrah & excitement of AI image generation is a bit past and the SHOCK VALUE of crap like the Taylor Swift stuff a year ago isn't going to be a normal thing.
I use several AI bots for coding, work, & other projects and Grok is the one I usually go to for a first try at a quick image I need.
Yes it was on Flux over Aurora but I expect the filtering will be OK in the end anyway
Completely agree - the hugely aggressive censorship of AI by risk-averse management is holding back the development and utility of the technology. I understand the reasoning of the other side, but I think it’s worth taking a bet on this approach.
DeleteI don’t think Grok is great for my use cases (rarely generating images) but it’s brilliant to get a slightly more ‘disruptive’ (in the truer sense of the word) competitor in the market alongside the establishment players who have quickly cornered it.
What is the utility of this technology on X exactly?
Deletethis can be so useful for video content creaters
DeleteImpressive. It performs better than DALL-E 3 and Google's offering.
ReplyDeleteMost photo realistic image generation model ever.
ReplyDeleteGrok 2 (beta) currently giving better results than Grok 2+ Aurora(beta); possibly with training will surpass it.
ReplyDeleteI guess there are limited number of people who have grok subscription
ReplyDeleteI have a regular X Premium and access to Grok.
DeleteI can see the option but I'm not able to use Aurora. It automatically reverts my selection back to grok 2 (beta). Wat da??! Anyone know why?
ReplyDeleteWas doing that on my Android App; desktop works fine
DeleteThis comment has been removed by a blog administrator.
ReplyDeleteits own
Delete
ReplyDeleteMaybe It's dream machine's image model, they have according to them they demonstrated substantial visual performance and prompt flowing gains while reducing the processing power required to generate images by an order of magnitude.
Edit I thought It was a dream machines model but I may have been mistaken I know It was the image model by one of the big video generator AI providers and It was revealed but not released ~1 week ago.
How much do you have to pay for unlimited gens?
ReplyDeleteThe question is:
ReplyDeleteDoes Mr. Heinlein approve the use of his Martian language?
There is no Stranger in a Strange Land than Stranger People in a Social Medialand
Is this a model from xAI or is this another open source model?
ReplyDeleteIt is a finetuned Version of Flux, nothing special.
ReplyDeleteIt appears it’s not https://x.com/eddib17_/status/1865485046943027481?s=46
DeleteThis comment has been removed by a blog administrator.
ReplyDeletehttps://imgur.com/a/zD1VFzr
ReplyDeletegenerated by grok today.
Bruh if OAI had made a Dalle with such "lax censorship" people would be on their throats...
DeleteBut Elon is doing it so there will be no consequences..
yeah, bold enough to persue AGI with non-profit camouflage, but intimidated to seek any truth of generated content. Don't know if hypocrites will lead to AGI😂
DeleteIt's more like people treat OAI harsher than any other AI company even before this whole non-profit debacle...
DeleteEveryone criticizes OAI for safety reasons but no one even brings up Google or xAI's safety efforts..
I've been following this circle for only a year and that's just my observation from a previous outsider's perspective.
Part of the reason is that OAI tried to court the safety crowd, and xAI never did.
DeleteThat was OAI's mistake. There is no satisfying the safety crowd. AGI's agentic nature is fundamentally incompatible with what that crowd wants. At best we can get some "alignment," but it's silly to worry about alignment until way later in the game, we are nowhere close to it being a real problem.
The other reason is that OAI may be attempting to engage in regulatory capture. Many times Altman has said something along the lines of "Our models are so dangerous that we are scared to release them. Please regulate us!" Of course you're gonna get more scrutiny if you say this, compared to a company that simply opens access to their models as they're developed.
IMO the way OpenAI has engaged in safety, alignment, and regulatory discussions has contributed to what is essentially a foot-gun moment where now everyone holds them to an unreasonably high bar. They really shouldn't have pretended they have some barely-restrained kraken internally and that only regulation can save the world. It was an incomprehensibly dumb PR move.
OAI is now valued at 150B, how is it treated harshly? I use LLM daily as coding copilot and could clearly see that Claude Sonnet is now much more capable than OAI models(whether it is 4o or o1) in terms of coding. I am quite sure now that OAI will be run into ground eventualy with Sam Altman as CEO, who played too much non-senses(like the one we discussed here) when OAI had an upper-hand.
DeleteWhat does valuation has to do with public treatment?
DeleteAnd again... What does being worse at coding have to do with being criticized for safety efforts?
Sam has been the CEO since the beginning...
What's this whole comment of yours? It feels like a mix of points that don't even match with each other...
coding is probably the most objective task where no censorship is needed, but OAI sucks at coding compared to claude and even worse than some Chinese LLMs😂 this tells you how much OAI has been downgraded technically while Sam is busy censoring their generated content. Now with Illya and co. gone, Sam will lead OAI to ground most likely.
Deletehey're no. 2 in coding... Well now no. 3 since Google's new model seems to show a lot of promise...
DeleteBut I wouldn't call it being bad at coding. And I don't want to hear anything about censorship from an Anthropic fanboy (I'm a former Claude user)
not a fan of anthropic, not even a subscriber of it. I just use claude sonnet daily on cursor. in terms of coding, obviously no censorship is needed. objectively speaking, gpt-4o is far worse than sonnet 3.5 and worse than deepseek v2.5. even o1 is worse at coding than sonnet 3.5. and recently, even github copilot support claude sonnet 3.5😂 this tells you how bad openai is in terms of coding AI.
DeleteI'm still wondering what all this has to do with the original comment of people treating OAI harsher for "safety" reason... You are trying so hard to side track the issue...
DeleteThat’s because Elon leaned into it, people expect it now, it’s not a shock. Also, realize with the censorship lax, the world hasn’t ended yet. OpenAI flinched at it, they took the other path, if something controversial is generated by dalle then people notice and it’s in a news article about how dangerous it is.
DeleteHad to try it, more photo-realistic than the previous one.
ReplyDeleteI'm testing Grok 2 Beta v Grok 2 + Aurora this morning with the exact same prompts Ive been using for months with the previous Groks versions. And the G2+A results are garbage compared with the original Grok 2 Beta and Grok 2 mini results. Like it's a huge step backwards in realism and just getting basic things right. G2+A produces lots of freakish people with multiple body parts and weird grey skin whereas G2 Beta produces very realistic people from the same prompts.
ReplyDeleteYou should probably start from scratch rather than feeding it an existing complex prompt that worked for a different model. Iteratively see if you can build up to your desired results.
DeleteAll of this is still a very nascent field. The way we do things today is extremely likely to be worthless in a year or two. It’s not like learning to drive today, it’s like learning to drive after the first car was invented.
DeleteOk, extra weird twist in the saga. This morning Grok 2 Mini was gone and Grok 2 + Aurora Beta and Grok 2 Beta were available. The results were horrible for me from Aurora but fine from regular G2 Beta.
DeleteThen a few hours later Groke 2 + Aurora and Grok 2 Beta are both gone and are replaced by Grok 2 + Flux Beta. The results from it from the same prompts seem fine. It does as few weird random things but the general results seem much closer to the G2 Beta and G2 mini results.
Anybody have a clue what is going on?
Same. Aura is atrocious. It draws people with three legs, six fingers, etc. And it prefers not to draw white people unless you specifically ask it to and even then it simply doesn't generate a photo half the time.
DeleteYeah,I only tested with people but I was getting amazing, lifelike results out of G2 beta and G2 mini before and then the same prompts in Aurora return mutant creatures with reptile looking skin and arms merging with heads and feet.
DeleteI'm guessing thats part of why it seems to have been pulled? I thumbs downed every image it generated. It even did a poor job with the backgrounds (way too simple) and lighting (way too even and "CGI" looking).
It's gone from the best photo generator to the absolute worst in a day. Canva's free built in draws about the same level.
DeleteI had it absolutely deny the fact that it could make images immediately after making an image. It didn't do it in a new session though.
ReplyDeleteThat's because grok is essentially acting as a middle man between your prompt and the image generator. Instead of pasting your prompt verbatim into flux, grok instead summarizes your prompt for flux which then generates the image and passes it through the censorship filter and then grok displays it.
DeleteIt's an awful system lol
Not available in the UK yet
ReplyDeleteit is now
DeleteThanks for mentioning this, I didn’t realize it and just saw it on Grok. Seems XAI will release some good updates soon?
ReplyDeletenot sure, do you know if they ever gave a general timeline for grok 3?
DeleteNothing confirmed but with the big investment announced recently and also xAI having the most GPUs, I would think there would be a big update soon. Possibly timed to make bigger headlines over Open AI’s 12 days thing
Deletei press fun mode and now i can't enable Aurora.
ReplyDeleteMine said I had Grok 2 and Grok 2 w/ Aurora. Now they changed it now says Grok 2 + FLUX or a Fun Mode toggle.
DeleteHas anyone had trouble analyzing images since this launched? I can create but it can't seem to analyze the uploads anymore (have Premium too).
ReplyDelete"I'm sorry, but I currently cannot analyze images because you've attached an image in this request"
Noice!!!
ReplyDeleteThat's Australian Intelligence speak for "nice". It's all in the way you say it.
Noice, give it a go.
It’s good. It’s really good. It follows directions so much better than other image generation models. It keeps multiple details active without them leaking into each other. I’m a fan.
ReplyDeletethey removed it why?
ReplyDelete