What are your thoughts on Sora?
- Thread starter Starbucks
- Start date
That’s what I’m aiming at, that we can’t differentiate the real with the generated photos.
Imagine in 5 years, people likely won’t be able to tell the difference anymore. Especially for those that believe already certain generated videos and images that are going viral on social media.
It’s quite dangerous and I’d love to see the EU coming up with some kind of label that obligates companies to auto add on AI content, including a burned-in meta data.
Imagine in 5 years, people likely won’t be able to tell the difference anymore. Especially for those that believe already certain generated videos and images that are going viral on social media.
It’s quite dangerous and I’d love to see the EU coming up with some kind of label that obligates companies to auto add on AI content, including a burned-in meta data.
Oh gawd, not more government intervention. They've done enough damage with the onerous Online Safety Act here in the UK. We really don't need more, but I'll bet we get it.
This is the first time I hear of Sora. For image generation I've been using Grok, but compared to what you are showing here, Starbucks, Sora makes Grok look like a doodler. Highly impressive!
I'll be keeping an eye on this, as I might use it once it becomes more publicly available.
I'll be keeping an eye on this, as I might use it once it becomes more publicly available.
It’s actually already available for every ChatGPT plus user! I already got access to it!
I was *this* close to getting ChatGPT Plus in order to have access to Sora. Upon reviewing the plans I've noticed that the Plus plan ($20/month) does not remove watermarks, only the Pro plan ($200/month) does. So I dug into what that would look like.
Through this image I landed onto this review.
Some interesting critiques on Sora to consider:
www.makeuseof.com
Especially with the watermarks I agree that this would be problematic. As shown in the above image, the watermark is placed quite a bit from the edge. If it would be all the way at the bottom, it would be easier to crop out (which I do with certain free services). OpenAI knows what it is doing.
I have a significant problem with this. Paying 20 bucks a month and still getting your material watermarked? No, sorry. The other points of critique in the article also seem valid. So I'll sit this one out. Hopefully in the future when Sora matures some more, the mentioned issues will be resolved, with most notably the removal of the watermark for the Plus plan.
Through this image I landed onto this review.
Some interesting critiques on Sora to consider:
My 9 Biggest Problems With OpenAI’s Sora
Sora makes magic—but only when the prompt behaves.
www.makeuseof.com
Especially with the watermarks I agree that this would be problematic. As shown in the above image, the watermark is placed quite a bit from the edge. If it would be all the way at the bottom, it would be easier to crop out (which I do with certain free services). OpenAI knows what it is doing.
I have a significant problem with this. Paying 20 bucks a month and still getting your material watermarked? No, sorry. The other points of critique in the article also seem valid. So I'll sit this one out. Hopefully in the future when Sora matures some more, the mentioned issues will be resolved, with most notably the removal of the watermark for the Plus plan.
Exactly, Retro. It's unacceptable.
Now that I have some more time on my hands I will look into the following alternatives:
Runway ML (Gen-2)
Starts free with watermark; $12–$28/month for plans that remove watermark and give more credits.
runwayml.com
Pika Labs
Currently free in Discord (still in early stages); watermark may appear depending on usage, but will likely have affordable paid plans when they launch commercially.
Kaiber
Starts at $10/month; no watermark at this tier. Apparently great for music videos and animations.
Hailuo
Starts at $108/year, no watermark. I already have some experience with Hailuo. Text-to-video or image-to video. But results vary.
hailuoai.video
Now that I have some more time on my hands I will look into the following alternatives:
Runway ML (Gen-2)
Starts free with watermark; $12–$28/month for plans that remove watermark and give more credits.
Runway | Tools for human imagination.
Runway is an applied research company shaping the next era of art, entertainment and human creativity.
runwayml.com
Pika Labs
Currently free in Discord (still in early stages); watermark may appear depending on usage, but will likely have affordable paid plans when they launch commercially.
Kaiber
Starts at $10/month; no watermark at this tier. Apparently great for music videos and animations.
Hailuo
Starts at $108/year, no watermark. I already have some experience with Hailuo. Text-to-video or image-to video. But results vary.
Hailuo AI: Transform Idea to Visual with AI
Hailuo AI Tools - the innovative ai video generator & prompt that transform your ideas into stunning ai videos. Create captivating visuals in no time with cutting-edge AI-powered tech and a piece of words. Unleash your creativity with Hailuo AI now.
hailuoai.video
Crikey, there's so many of them now, I've never heard of the others. Once you've compared and chosen one, do you think you might be able to do a mini review of it comparing it to the others that you've tried? I'm not suggesting anything really long and detailed that would take a of time and effort to do, just a quick post to give us some idea of your experiences with them. 
Yeah sure. I'm having a Hitcore project with animations in mind, so I will be playing with these things anyway on their free plans. I'll share some of the output (nevermind the watermarks for this purpose then) so we can compare the quality.
Then with the best one I will proceed to create my projects.
Then with the best one I will proceed to create my projects.
Alright, in the intro of my video I want to show, in anime style, a digital alarm clock at 4:59 jumping to 5:00. Simple enough, right?
For each generated video I used the same prompt 4 times. Out of those, below the best results:
Runway: https://app.runwayml.com/creation/d632ea5c-12ac-461c-9d73-ff2f42fb7322
Pro: Graphics are pretty much how I envisioned the clip. Nice.
Con: Doesn't understand something as simple as numbers. It displays nonsense, every time, also with variations in the prompt,
Pika: https://pika.art/video/d9b55536-6a8c-4945-9aaf-4e14adca3239
Pro: It's trying, I guess??
Con: It makes a mishmash of things. The numbers are off. Also, the alarm clock seems to be a digital/analog hybrid.
Kaiber: really don't like the tedious interface. Skipped.
Hailuo: https://hailuoai.video/share/ai-video/zKJgEvpZK4yb
Pro: Fairly decent image, after a few attempts and adjustments
Con: Much like Runway it is messing up the digits.
Conclusion:
Pika and Kaiber are out of the race. Runway and Hailuo have potential, but some bugs need to be fixed. I'm willing to give them a second chance at some point. For now I am trying to look for other alternatives.
For each generated video I used the same prompt 4 times. Out of those, below the best results:
Runway: https://app.runwayml.com/creation/d632ea5c-12ac-461c-9d73-ff2f42fb7322
Pro: Graphics are pretty much how I envisioned the clip. Nice.
Con: Doesn't understand something as simple as numbers. It displays nonsense, every time, also with variations in the prompt,
Pika: https://pika.art/video/d9b55536-6a8c-4945-9aaf-4e14adca3239
Pro: It's trying, I guess??
Con: It makes a mishmash of things. The numbers are off. Also, the alarm clock seems to be a digital/analog hybrid.
Kaiber: really don't like the tedious interface. Skipped.
Hailuo: https://hailuoai.video/share/ai-video/zKJgEvpZK4yb
Pro: Fairly decent image, after a few attempts and adjustments
Con: Much like Runway it is messing up the digits.
Conclusion:
Pika and Kaiber are out of the race. Runway and Hailuo have potential, but some bugs need to be fixed. I'm willing to give them a second chance at some point. For now I am trying to look for other alternatives.
Those glitches that AI generates in your examples are why I don't trust AI Google results, since AI of all brands can hallucinate and is a known problem with the current generation. If I search for something, I want the actual source, ie a standard website result that I can evaluate and see if I think its provinence is good enough to trust, like the NHS for medical queries, for example.
Having an AI result to complement it is useful and I like that it's there in many search results, but it shouldn't replace the sources that generated that AI result in the first place.
