ChatGPT announces Sora, a tool for text to video. Type in a prompt and generate to create a video. This may be good for us in generating pregnancy AI. It has not yet been released, can’t wait.

I mean if we thought Dalle3's censorship was taking the piss, you know rightly OpenAI will clamp down even harder for video.

Quote:For example, once in an OpenAI product, our text classifier will check and reject text input prompts that are in violation of our usage policies, like those that request extreme violence, sexual content, hateful imagery, celebrity likeness, or the IP of others. We’ve also developed robust image classifiers that are used to review the frames of every video generated to help ensure that it adheres to our usage policies, before it’s shown to the user.

What it will do is force competitor's like StabilityAI to push even further to producing good T2V models (Meta already pushed out some new training code), like SVD will be pretty good (ok not Sora level, but that's another bar to each) once they actually release the T2V component rather than just I2V without controls.

Can only imagine the amount of compute required to locally run Sora-Level video generations, might need to wait another RTX generation or two.

Animatediff now has easily trainable Motion Lora's, which from my testing might allow us to get some interesting pregnancy generations....but it's still in the experimental phase.

Don't get me wrong it looks fucking class, game changer, but it's very much for wholesome (and sometimes cursed ) content.

Case anyone hasn't seen the report:
hxxps://openai.com/sora

Some examples from the technical report (randomly jumps through them), if we can trick the censorship when it checks frames this could be magical.

http://tinyurl.com/yck9j3ve
http://tinyurl.com/v7xeepnw
http://tinyurl.com/uhbr5rjp
http://tinyurl.com/yc4py7vz
http://tinyurl.com/4e3np4st
http://tinyurl.com/2p8sta2r

The technical report:
hxxps://openai.com/research/video-generation-models-as-world-simulators

doesnt chatgpt censor a lot of content though?

You can trick chatGPT fairly easily, depending on the content you want.

Let's just bare in mind that Bing Image Creator/Designer is effectively a wrapper of Dalle3, yes it has more censorship than using it directly via ChatGPT, but Dalle3 is still pretty strict.

The issue isn't the prompt being blocked, but the generation itself being blocked if it detects anything dodgy in the image generated, and for video that'll have to be stricter. You can't trick it when you don't know what is being generated behind the scenes.

That's been one of the bigger problems with Dalle3, even for wholesome as fuck prompts, sometimes the generation might contain something from it's list of blocked concepts that you cannot account for.

For a 60 second video it'll perform that check on each and every frame, so it reduces the chances even further of a successful generation. But hey maybe it'll be happy with conservative fully covered bellies.... wonder will it even allow sports women or beaches XD

I've been trying to do it but it's not working... is it out yet?

The outputs that have been released really are remarkable. This will transform porn/adult content if people are able to get around restrictions.

I'll always appreciate authentic/real women, for whatever reason I find it more arousing (perhaps the notion that she's there and, in theory, I could bang her myself), but I'm unsure as to how common this is. Webcam girls are here to stay, in large part because of the interactive/parasocial component, but I do wonder how much conventional porn will be replaced. The stuff that's primarily or purely meant to serve as a visual stimulus.

I just got an email to join a waitlist for access to Sora. Has anyone else had any luck?