Avid Reading Horse
Member
Where do codes come from, aside from GAF benefactors? They never send me shit despite being a paid $20/mo customer for over 2 years 
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
Sora 2
- Thread starter Sybrix
- Start date
RedC
Member
What's the original prompts for those?
12Goblins
Banned
I can't imagine having to clean up code. Sounds like a nightmare...AI is Industry creating
ResurrectedContrarian
Suffers with mild autism
It's creepy... this seems like it could really be him, as do a lot of the celebrity recreations.
exactly like the holodeck on star trek, I though it was a hundred years away before we'd be able to conjure up comedians or anything and have them seem real
don't know -- remixes of remixes, which is how a lot of things on Sora move now (a fun concept actually... someone gets a "hit" and others swap the characters etc)
Last edited:
Sokka
Member
Love the portrayal of Zoro. just wish Fletcher would swear some more.
Avid Reading Horse
Member
RedC
Member
Yeah, I know. You can still see the original from which they remixed it.
I really like the Aerith and Rinoa videos.
ResurrectedContrarian
Suffers with mild autism
but the original changed the label to just "Aerith"... when you do that, doesn't it effectively hide your real prompt? I don't see a way to get back to it, I only see their updated label.
RedC
Member
Last edited:
kpopSuperstar
Member
I still havent seen anything but memes and fun clips, which is weird, there should be something more serious at this point.
Draugoth
Gold Member
Last edited:
Avid Reading Horse
Member
Couldn't get it to render Phil by name but I am reasonably happy with this, brother!
Last edited:
falconhoof
Member
Sora update #1
We have been learning quickly from how people are using Sora and taking feedback from users, rightsholders, and other interested groups. We of course spent a lot of time discussing this before...
blog.samaltman.com
Better be quick making IP videos
Also,
AREYOUOKAY?
Member
There goes all the fun.Sora update #1
We have been learning quickly from how people are using Sora and taking feedback from users, rightsholders, and other interested groups. We of course spent a lot of time discussing this before...
Better be quick making IP videos
Also,
IntentionalPun
Ask me about my wife's perfect butthole
A woman flies over a lush landscape of various colored trees and purple swirling water. Her wings made of human flesh catch the sunlight, their veiny insides visible. All is well in the world as she lands atop her castle built into the hillside.
Avid Reading Horse
Member
Monkey donut saga:
kpopSuperstar
Member
It looks good and all, and I have seen plenty of trailers like this, but it's all short clips edited together, I have not seen a proper scene with a "person" that feels like a movie with a proper actor.
I think a full on AI movie will look worse than people think, it can look real and photorealistic, but I don't know if it can or will be able to fully come off as human in long cohesive scenes.
RJMacready73
Simps for Amouranth
Anytime I use these AI to try and generate realistic images of Judge Dredd scenes in mega city 1 I get nowhere near the fidelity shown in the promos, guess my AI-FU is off
RedC
Member
It won't generate a video because the content may violate their guardrails around nudity, sexuality, or erotic content.
It's likely the wings made of human flesh and the veiny insides visible, causing the issue.
To make it safe, I polished it up to:
A woman soars gracefully over a lush fantasy landscape filled with trees of many colors and rivers of purple swirling water. Her wings, uncanny and organic, catch the sunlight, glowing with translucent detail. All is calm and dreamlike as she descends and lands atop her grand castle built into the green hillside
IntentionalPun Request 1
IntentionalPun Request 2
I added camera edits for the following:
IntentionalPun Request 3
Last edited:
IlGialloMondadori
Member
I'm not sure what it is, it's hard to put a finger on it. Nothing in this scene is interesting, or exciting, or compelling, it's just like...images. There's an emptiness in AI generated videos that is always there, and a sheen. There's never a sense of "happy accident" perfection, or perfection of performance and editing and cinematography, it's just like, stuff.
I don't know if it's just because the people using it are terrible, or if it's just impossible.
I've never seen anything that comes close to feeling like a short film. It never feels alive.
E-Cat
Member
IntentionalPun
Ask me about my wife's perfect butthole
Hey I aooreciate it. It's not quite what h was going for of course, I wanted something that was a conbo of what it provides with an edge of creepiness to it. My creepy flesh wing request of course ruined it lol
It does feel a bit like it's more restrained by reality than image generators (even ignoring the censorship.)
Last edited:
IntentionalPun
Ask me about my wife's perfect butthole
Sick! Purple water… kinda creepy wings on that last one, more fantastical colors….very close.
RedC
Member
RedLetterMedia gets flagged by sora 2, however one time it did sneak through and generated this.
Trilobit request
RedC
Member
The prompt I used on the last one was:
A woman soars over a fantasy landscape of multicolored trees and swirling purple rivers. The camera sweeps wide, but the valley is unnervingly silent, the colors too vivid to feel natural. Close-ups reveal her wings, disturbingly organic, their translucent veins glowing in the sunlight like something alive. The smooth tracking shot follows her descent as shadows lengthen across the land. She lands atop her castle carved into the hillside, its spires jagged and looming, the structure watching her like a living thing.
Last edited:
Draugoth
Gold Member
AI needs Training Data and can't create or think. Most of these videos are heavily overfitted.
Because LMs use Tokens for memory context, if you prompt something too long it will degenerate into an incomprihensible mess. You can notice this on long ChatGPT conversarions, the bigger they are, the worse they get.
Trilobit
Absolutely Cozy
That's hilarious! Like some sort of alternate universe RLM.
What makes it flag some stuff and not others? Like why does it allow Robin Williams so well? I've never tried Sora, but maybe in the future.
RedC
Member
More things are getting flagged as we speak, so a lot of videos I made in the past with celebrities and copyrighted IP, I can't make now.That's hilarious! Like some sort of alternate universe RLM.
What makes it flag some stuff and not others? Like why does it allow Robin Williams so well? I've never tried Sora, but maybe in the future.
Last edited:
Draugoth
Gold Member
Not enough Training Data. It's only two movies, meanwhile South Park has 330 episodes, One Piece 1000+ Episodes
Also it was confirmed the Demo Clips in the release video were altered by humans
Last edited:
IntentionalPun
Ask me about my wife's perfect butthole
Nice work.
I use generative AI for software development and it's all about giving the system loads of context.
The company I'm working for…uhhh… makes one of the tools, and internally they have servers that provide loads of context and instructions for all kinds of things but mostly software development.
What I'd love to see if Sora can do, but I doubt it can, is produce some good acting:
I know it's only short clips for now (that cost a few hours of using a microwave's energy lol) but I do wonder if there's a prompt that would produce some dramatic acting that we could see if it's any good.i
I'm on the side of all of this being impressive but kind of empty and shallow. The way people talk in particular doesn't feel genuine.
Last edited:
Draugoth
Gold Member
It was clearly a marketing stunt to get ahead of competitors like Veo and Wan, it worked but it seems to have backfired. I wouldn't be surprised if lawsuits started rolling in. Now people are aware that these models did indeed stole from everywhere basically.
Last edited:
RedC
Member
RedC
Member
Draugoth
Gold Member
Yeah, I think we've hit scaling walls like we did with Image Diffusion models last year. There's only so much you can do with training data available, I can see the weird texture glitching in lots of these videos. Idk what they are but seem to be caused by the AI generating each single frame separately while the most impressive are overfitted
RedC
Member
Heimdall_Xtreme
Hermen Hulst Fanclub's #1 Member
Is it already included in ChatGPT plus or is it separate?
RedC
Member
It's included, but you have to redeem an invite code as well for this latest version. Which I can share with you.
100% a frame by frame issue. I recognize that 'look' immediately lol. It's trying to figure out lighting and color settings from scratch for every individual frame.
Current AI models will never be more than a way to make some goofs and gafs. We need something like AGI to be trained on nuance, context, timing, tempo etc. None of which the current models do well unfortunately.
That car vid you linked is a good example. That would never ever fly for an actual company since they cant afford their product to look slightly different in every shot. Not to mention all the recompositing you would have to do for all the text. That alone would take hundreds of hours. Might as well take a couple of drones and action cams to a race court at that point.
Heimdall_Xtreme
Hermen Hulst Fanclub's #1 Member
If it's not too much trouble, I'd appreciate it.
.
RedC
Member
Avid Reading Horse
Member
Definitely running into struggles with marginally more creative concepts... took a lot of renders to get something that was kind of what I was going for:
Never got a satisfying break. Never got the football players mouth to move so I guess ya just have to treat it like an internal monologue. One of the renders was total trash but did quite clearly play the Curb Your Enthusiasm ending credits song for a couple of notes.
Revolutionary War except we have an F18:
Made this one for a buddy and thought it turned out well. When I hyped up LLMs to him when they came out he wanted to use it to find old TV shows he couldn't quite remember. One of them was something about talking claymation cars. ChatGPT never gave him an answer he was happy with. I think he's misremembering it. But I rendered some concepts when DALLE was the thing... and then I had ChatGPT look at the picture and make a prompt for an animated version.
Never got a satisfying break. Never got the football players mouth to move so I guess ya just have to treat it like an internal monologue. One of the renders was total trash but did quite clearly play the Curb Your Enthusiasm ending credits song for a couple of notes.
Revolutionary War except we have an F18:
Made this one for a buddy and thought it turned out well. When I hyped up LLMs to him when they came out he wanted to use it to find old TV shows he couldn't quite remember. One of them was something about talking claymation cars. ChatGPT never gave him an answer he was happy with. I think he's misremembering it. But I rendered some concepts when DALLE was the thing... and then I had ChatGPT look at the picture and make a prompt for an animated version.
Draugoth
Gold Member
I gave up on Nano Banana when it was unable to clean a Silksong poster for me, so I went back to Photoshop and got better results.
The reason the Copyrighted stuff looks so good it's because they were both overfitted and heavily popular, with a lot of data to push from the training.
We have tens of thousands of police cam footages as well.
Anything that was never created and only exists in your head? Good luck with that, at best you will have a good prototype for your project. The AI companies have already run out of data to train.
I'm not kidding. The best and biggest usage for Generative AI in the future in my opinion will be Porn.
Last edited: