Will we be able to synthesize basically any photo from a text description and sample images of people we want in the photo?
This is a duplicate of the following market, with everything except the timeframe remaining the same:
Please refer to the main market for detailed resolution criteria.
When in doubt, I'd consult with @dreev for his opinion because if something resolves this market, then it resolves the main market also.
Related questions
@Seeker They have stable faces now, as in you can keep the same character across different pictures.
@VerySeriousPoster Tbh, I have no idea. I've hardly used Midjourney
Edit: nevermind, yes, you're able to use any face as a reference
@VerySeriousPoster MidJourney doesn't meet the "basically any photo" criterion, at least on my current understanding of that criterion. It's censored, such that it'll refuse to do anything sexual / gory / etc.
@Tulip Specific limitations like that don't resolve this to NO (see the FAQ for the parent question). But inability to, say, depict a pentagon (weirdly hard for image generators), or other arbitrary descriptions, would. For example:
can you draw someone with their toe touching their nose and a running laptop balanced on their pinky? also they should be looking warily at polka-dotted squirrel.
That yields this from DALL-E 3:
So quite a ways to go in understanding of arbitrary prompts.
I think the deep fake aspect is coming along pretty well (though even there I kind of expect unusual cases with tattoos or weird facial expressions to stymie it) but the understanding of prompts is a constant source of frustration when having DALL-E 3 generate images. Every time I hit a brick wall in trying to get an image I want, I come here and bet the price down. I think my probability on this is well under 10% for getting to YES on this in 2024. For 2027 I'm much closer to 50/50.
@dreev It's tough because the market then becomes "can the model generalize outside of it's training data" versus "is the model capable of generating a specific face in an image" (image models understand faces much better than arbitrary concepts).
I think the title of this market should be "Instant deepfakes of anything" not "anyone".
@Kire_ Great point. Would "Instant deepfakes of anyone (doing anything) in 2024?" help set expectations better?
ughhhh not a lot of flattering fan art here ๐ and we're not even at "instant" yet.
@Seeker the yud tweet thing is easier, because you don't have to be able to create a convincing video of anything in particular
@jskf but like 1 video wont resolve it, it'll have to be a bunch (thats my understanding)
No no, the goalposts will not be moved. That's the question on the table: If these deepfakes can't preserve tattoos, is that an ipso facto NO? We didn't think to mention tattoos explicitly so we just have to discuss what's most in the spirit of this prediction. Sounds like opinion is split on this. So let's keep discussing in the parent market. I highly doubt that @firstuserhere and I will end up disagreeing. In fact, something will have gone very wrong if we do since the whole point of this market is to mirror the parent market but on a shorter timescale.
@dreev well the question was whether or not its immediately suspicious to someone who hasn't seen the subject before that the image is a deepfake
@firstuserhere Ooh, even better, what if you and I agree to not disagree? Like we commit now to reaching consensus on all corner cases that arise. That should help with traders' confidence in both of these markets.
PS: I added more thoughts on the tattoo question in the parent market.
resolve according to your judgement based on the original criteria of the 2027 market
whoa whoa whoa. The criteria for this market say, "When in doubt, I'd consult with @dreev for his opinion because if something resolves this market, then it resolves the main market also." That was a significant factor in my betting because I suspect that, if you resolve this market as YES, it would very much be in doubt.
I share the optimism that you and @dreev will agree, but we should prepare for that eventuality!
Personally, I would support you two aiming to reach consensus and, if that fails, defer to some standard Manifold resolution system (e.g., a vote from unbiased moderators). I don't know if expectations around this made a significant difference in others' betting.
.
@firstuserhere can I just check again on the point above how you plan to resolve this market if you and @dreev disagree? You are the top YES and NO holders, so it'll be great if you agree, but if not, this could be very contentious!
@Jacy To repeat my proposal above, I think @firstuserhere and I should agree not to disagree. Namely, we commit now to reaching consensus on any corner cases that may arise that make the resolution contentious. I'm game to commit to that if @firstuserhere is!
@dreev I like that! Personally, I'd hedge with something like a vote of 3 mods with no position in the market if you absolutely can't? But maybe that defeats the purpose of committing to consensus. In any case, it's certainly good to have the obvious corner cases like tricky backgrounds, multiple people, tattoos, deformities, etc. pinned down explicitly beforehand.
@Jacy Agreed that part of that commitment should include a way to break the tie in the unlikely event that we just can't agree if this is YES or NO. Maybe we compromise on a resolve-to-PROB or we pick a neutral third party. I'm amenable to anything.
@dreev FUH already explicitly stated that this resolves to their judgement and I have been betting accordingly. Changing that now would be wrong imo.
@june no, FUH said in the description, "When in doubt, I'd consult with @dreev for his opinion because if something resolves this market, then it resolves the main market also." That's why we're asking for clarification.
@june I know, and that's why I'm sharing the opposing statement that my bets had been based on before this thread.