SoylentNews is people
SoylentNews
On Tuesday, Stability AI released Stable Video Diffusion, a new free AI research tool that can turn any still image into a short video—with mixed results. It's an open-weights preview of two AI models that use a technique called image-to-video, and it can run locally on a machine with an Nvidia GPU.
Last year, Stability AI made waves with the release of Stable Diffusion, an "open weights" image synthesis model that kick started a wave of open image synthesis and inspired a large community of hobbyists that have built off the technology with their own custom fine-tunings. Now Stability wants to do the same with AI video synthesis, although the tech is still in its infancy.
[...]
In our local testing, a 14-frame generation took about 30 minutes to create on an Nvidia RTX 3060 graphics card, but users can experiment with running the models much faster on the cloud through services like Hugging Face and Replicate (some of which you may need to pay for). In our experiments, the generated animation typically keeps a portion of the scene static and adds panning and zooming effects or animates smoke or fire. People depicted in photos often do not move, although we did get one Getty image of Steve Wozniak to slightly come to life.
Previously on SoylentNews:
Search: Stable Diffusion on SoylentNews.
(Score: 2) by looorg on Friday December 01 2023, @03:48PM (4 children)
This is my personal main grief about it all. There is no proper quotation or citation in any of it. Just vague references to "science", knowledge, papers or reports or that someone or things or wrote or said something. But not who, or links or what it is really basing any of its conclusions on. This make almost anything it produces worthless and more of a novelty. Without proper citation it's a novelty item and not actually useful since it become so hard to check the validity of it all.
It can make text, it can make interesting and pretty pictures (and some really disturbing once). But beyond that it's of little actual usable value.
The best way you can work with it then is to limit the data it can work with where you know where it all came from. Trying to find hidden patterns in data and such things. But that requires a lot of manual work from you to set it up correctly. Still checking the results is very tedious and time consuming. But it also somewhat defeats the purpose of it all if you have to spend all that time properly curating data for it.
(Score: 2) by Freeman on Friday December 01 2023, @04:23PM
It's rather good at translating a bit I need to say in an email into "corporate speak" so I don't need to use valuable brain cells to convert my thoughts into "sounding good". It's easy enough to proof read and edit a good enough result.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 0) by Anonymous Coward on Saturday December 02 2023, @12:21AM (2 children)
> There is no proper quotation or citation in any of it.
Speculation: The developers were too cheap to buy access to the masses of scientific and engineering papers that are behind paywalls. So most of the research papers used in the training set is just abstracts.
(Score: 2) by aafcac on Sunday December 03 2023, @06:55AM
Or they couldn't. There's lots of issues when it comes to letting ML programs take information from various articles and synthesize them beyond just payment. Plagiarism and incorrect conclusions are a real issue.
(Score: 2) by looorg on Sunday December 03 2023, @11:59AM
That is probably part of it. All these gigantic files of "training" data they scraped is probably not paid for properly. This is why we are seeing all these lawsuits and artists and rights-holders of various products are suing. They basically just copied everything thinking it was online so it was all free. Or they just didn't bother to make not of whom or what they copied. After all that is extra data that had little value to them. Or the AI at the moment just can't or won't share credits. Either it can't or it won't, it might not understand the concept or have the programming to be able to say that "this piece" comes from here so you can check it's work or conclusions. It might not want that since it would probably reveal the hallucinations. So no need to gather it. When and if that changes they'll just gather new data with that included.
I think it's a missed gathering criteria tho as if you knew where things came from you could easier remove all the bad data that has been gathered that may or may not be causing some of the hallucinating. But then as noted if they mentioned where things came from they might have to pay for it. They certainly didn't want that.
Basically as it stands to day the entire field is built on stolen data. One day they'll have to sort that out. It was painful for other companies that did the same things, some never managed to claw their way out of that dilemma. Others like Spotify spent a lot of effort removing all those illegally downloaded MP3 compressed audiofiles they built their service on. Here it might be worse since they might not know, it might be better to just scrap it all and starts scraping with purpose again.