AI Videos we have generated

[Brian Hall]

I didn't see a previous post on this and thought it was worth a thread of it's own. For any videos posted, please list the tools used and whether it was done with local AI tools or with a commercial generator, like Gemini, Grok or something else.

I generated this short clip with LTX 2.3 running in WanGP locally on my pc. The starting image was generated with Krea 2 running in Forge-Neo.

 

[Brian Hall]

This one was generated with Grok's Imagine. I uploaded a starting image I created with Krea 2 in Forge-Neo. I tried to generate the video locally, but LTX 2.3 has a problem with ballet movements.

 

[Brian Hall]

This one was generated in LTX 2.3 running in WanGP on my pc. The starting image was generated with Krea 2 running in Forge-Neo, also on my pc. I was trying out a dark scene with a candle.

 

[Brian Hall]

I had Grok generate a better video for the idea I had for the ghost ballet and this second video has inspired some new lyric ideas for a new song. This is the second and better video:

 

[Pauli]

Just my 2 ¢s:

With focus on the technical aspect one may find the PC-based single-user generation of AI videos quite interesting. Nevertheless, in a broader view questions come to mind like »Why in the world should one want to generate AI videos at all?« or »What content (or even message) will typically be implemented – and could that make the resulting videos really worth watching?«

 

[sergeauckland]

Just my 2 ¢s:

With focus on the technical aspect one may find the PC-based single-user generation of AI videos quite interesting. Nevertheless, in a broader view questions come to mind like »Why in the world should one want to generate AI videos at all?« or »What content (or even message) will typically be implemented – and could that make the resulting videos really worth watching?«

I can see Advertising Agencies using this, as a lot cheaper than hiring actors, venues, video editors etc etc. It must be worrying for the vast number of actors who need this sort of work to make a living. Ditto composers of jingles, backing music, and anything creative that doesn't rely on the name of the performer for its value.

S.

 

[Pauli]

I can see Advertising Agencies using this

Sure. But here we are dealing (I guess) with a more amateurish PC-based single-user scenario instead.

 

[Brian Hall]

Just my 2 ¢s:

With focus on the technical aspect one may find the PC-based single-user generation of AI videos quite interesting. Nevertheless, in a broader view questions come to mind like »Why in the world should one want to generate AI videos at all?« or »What content (or even message) will typically be implemented – and could that make the resulting videos really worth watching?«

To learn something new. complicated and interesting. To bring ideas to life without hiring an expensive film crew and actors. It isn't great yet, but the local non-commercial models and tools are advancing quickly. 6 months ago, most images and videos produced by AI models always had messed up eyes, fingers, facial expressions, etc. It still happens, but is becoming more rare.

As for whether the resulting videos are worth watching, if you aren't interested, just ignore them. I'm retired and have plenty of time so I'm learning how this stuff works like another hobby.

 

[Brian Hall]

The new LTX 2.5 video model released yesterday and WanGP released an update to support it today. I generated this short clip with a starting image of Mona Lisa and added one of my original songs for her to sing.

 

[kemmler3D]

I can see Advertising Agencies using this, as a lot cheaper than hiring actors, venues, video editors etc etc. It must be worrying for the vast number of actors who need this sort of work to make a living. Ditto composers of jingles, backing music, and anything creative that doesn't rely on the name of the performer for its value.

S.

I work in the advertising space now, and yes this is a pretty relevant use. Take a picture of a product, make it into a video for TikTok ads or whatever. It will definitely eat up some of the low end of the market for actors / videographers.

However, IMO even the cutting edge stuff is not quite ready for primetime in the sense of replacing a real video shoot. Quick short clips perhaps but longer or more involved stuff, no. Brands that don't want to be seen slinging AI slop will still do it the old-fashioned way, at least for now.

 

[Brian Hall]

I work in the advertising space now, and yes this is a pretty relevant use. Take a picture of a product, make it into a video for TikTok ads or whatever. It will definitely eat up some of the low end of the market for actors / videographers.

However, IMO even the cutting edge stuff is not quite ready for primetime in the sense of replacing a real video shoot. Quick short clips perhaps but longer or more involved stuff, no. Brands that don't want to be seen slinging AI slop will still do it the old-fashioned way, at least for now.

The AI models are evolving rapidly. I've been playing around with that new model today and it is very impressive. A huge improvement over the 2.3 version that came out earlier this year and it can run on a decently powerful pc with no need to pay for an online video generator. The license for it says it is 100% free unless you make over ten million dollars with a video generated with it.

 

[kemmler3D]

The AI models are evolving rapidly. I've been playing around with that new model today and it is very impressive. A huge improvement over the 2.3 version that came out earlier this year and it can run on a decently powerful pc with no need to pay for an online video generator. The license for it says it is 100% free unless you make over ten million dollars with a video generated with it.

The last one I messed with was the most recent Google release, and it was pretty impressive but also made mistakes of the kind that are unsuitable for ad use, in my case it took a right-handed circular saw and made it left handed, and also added a battery where there was none before. The realism of any given element was great in isolation, but the fidelity to visual inputs was not there. Not sure how LTX or whatever is in that regard.

 

[Brian Hall]

The last one I messed with was the most recent Google release, and it was pretty impressive but also made mistakes of the kind that are unsuitable for ad use, in my case it took a right-handed circular saw and made it left handed, and also added a battery where there was none before. The realism of any given element was great in isolation, but the fidelity to visual inputs was not there. Not sure how LTX or whatever is in that regard.

For things like that I would use a real picture of the saw as a starting or reference image.

 

[Brian Hall]

Video generated with LTX 2.5 running in WanGP using a starting image generated in Krea 2 running in Forge-Neo. Both were done on my pc.

 

[kemmler3D]

For things like that I would use a real picture of the saw as a starting or reference image.

I used several, it had pictures of the saw from at least 4 angles.

 

[Brian Hall]

I've generated some videos that I'm leary of posting on youtube because they might get a copyright strike for using generated images of real actors. I used a start image of Rambo I generated with Krea 2 in Forge-Neo. This video was generated with that start image using the new LTX 2.5 video model in WanGP.

Rambo Scene (0:00 - 0:16.9).mp4

drive.google.com

This link should allow anyone who is interested to view the 16.9 second video in Google Drive's player or allow it to be downloaded.

 

[Brian Hall]

A short video clip of a character I generated with Krea 2 in Forge-Neo. I used that as a starting image in the new LTX 2.5 running in WanGP. I attached an audio clip of one of my original songs for the character to sing. I prompted for wind blowing her hair.

This is the image I used to generate the video:

 

[Brian Hall]

A short video clip of a character I generated with Krea 2 in Forge-Neo. I used that as a starting image in the new LTX 2.5 running in WanGP. I attached an audio clip of one of my original songs for the character to sing. I prompted for wind blowing her hair. I've been learning how to generate a consistent person/face with the image generator. This clip uses the same or close face as the Visions clip above with different hair and clothes.

 

[Brian Hall]

A short video I generated with the new Minimax H3 video model in WanGP on my pc. The starting image used to generate the video was generated in Krea 2 in Forge-Neo.

 

[Brian Hall]

A character from the game Cyberpunk 2077, Judy Alvarez, eating an ice cream cone. Starting image generated with Krea 2 in Forge-Neo. Video generated with LTX 2.5 running in ComfyUI.