"Firefly for Video" will generate sound effects and music, edit footage via text.
See full article...
See full article...
Adobe teases generative AI video tools
- Thread starter JournalBot
- Start date
Remember when the only thing A.I. was, in the public's mind, just an unfinished Kubrick project that was "Spielbergified" into an "ok" movie? Those were good times.
Upvote
16
(20
/
-4)
ColdWetDog
Ars Legatus Legionis
Oh you sweet summer's child. You ain't seen nothin' yet.
Upvote
50
(50
/
0)
Upvote
12
(13
/
-1)
I know that a few months back a number of artist's found that Abobe had auto-checked "On" a box for:
Content analysis
Adobe may analyze your Creative Cloud or Document Cloud content to provide product features and improve and develop our products and services. Creative Cloud and Document Cloud content include but aren't limited to image, audio, video, text or document files, and associated data. Adobe performs content analysis only on content processed or stored on Adobe's servers; we don't analyze content processed or stored locally on your device.
Allow my content to be analyzed by Adobe for product improvement and development purposes
Adobe Content Analysis FAQ Page
Content analysis
Adobe may analyze your Creative Cloud or Document Cloud content to provide product features and improve and develop our products and services. Creative Cloud and Document Cloud content include but aren't limited to image, audio, video, text or document files, and associated data. Adobe performs content analysis only on content processed or stored on Adobe's servers; we don't analyze content processed or stored locally on your device.
Allow my content to be analyzed by Adobe for product improvement and development purposes
Adobe Content Analysis FAQ Page
Upvote
44
(47
/
-3)
Same. I'm very much for AI being a helper, especially with my coding.
But I do wonder how the more limited mindset of AI might wind up being similar to autotune or pop music all sounding the same. Of course, we managed that with just human imagination. Who knows, maybe it'll turn out that AI isn't as hackneyed as humans. I do wonder, though.
Upvote
23
(24
/
-1)
I was here before Rick Rolls. I was here before the first felid tentatively requested a cheeseburger in broken English, and a walrus demanded its bucket. I’ve drawn Kilroy in ink and pencil and cut it in negative topiary space. These things are all the product of human intelligence, mundane in form yet exceptional in how quickly and easily they transmit.
Will Smith eating spaghetti though, that is chaos walking. That’s a kernel of thought from no one, that accreted on the edges of common human experience the way shit rimes on the walls of a port-a-john at a winter carnival or Ohio State sporting event.
I’m a summer child? Winter is coming. Winter is here. We’re all going to die smothered in the sum of human creation devoid of context or limitation, proliferating out by endless witless imperative. That is what Will Smith eating spaghetti means. It means the end, and the death.
I am Samus. Samus is my name.
Last edited:
Upvote
55
(58
/
-3)
But even a thing like autotune was used creatively and uniquely (at first). Being popular so that everyone else copied and then built upon it is just part of the human creative process. A continually training AI should be able to ride the fads as well. That will give future historians an ability to pinpoint moments in time like the 2010s became the decade of the river table (in woodworking).
Upvote
4
(6
/
-2)
What I would like, what Adobe is not much into, is a voice to generative audio beats for musicians. Just talk into your mic, "give me beat". "No, slower". "Increase the volume of the hi-hats". "Create a changing measure every forth beat". "Make some rolls".
Something for the rhythmically challenged layman. Someone who just wants a beat right now without having to click a mouse to make dots on the screen. Something you could talk to to make the perfect beat that's in your head so you can just jam. Something even usable when you're sauced up. As long as you can talk, you can get the prefect beat.
Something for the rhythmically challenged layman. Someone who just wants a beat right now without having to click a mouse to make dots on the screen. Something you could talk to to make the perfect beat that's in your head so you can just jam. Something even usable when you're sauced up. As long as you can talk, you can get the prefect beat.
Upvote
7
(11
/
-4)
I've used Beat Map in Reason with some success that does what you want (I mean, it is knobs, instead of voice, but it has everything you wanted) https://reasonstudios.com/products/beat-map/
Some randomness in there, and then tweaking hi-hats, kick, snare, perc elements, and even keying them to increase or decrease density when you want.
Upvote
4
(4
/
0)
The visual effects and captioning seem fairly flawless in the demo reel, but I'm not a photographer or video artist. I am a musician, and to my ear the soundtrack falls straight into the uncanny valley, just as every other AI tool I've run into so far does. Some of it was just elevator quality, but that Mothra gospel choir at the end of the clip is the sonic equivalent of spaghetti Will Smith.
Adobe has access to top-notch critical and creative talent, so the fact that they're putting this weirdness out there tells me that they don't think the difference in production quality is noticeable to the average listener. And that's disconcerting.
Adobe has access to top-notch critical and creative talent, so the fact that they're putting this weirdness out there tells me that they don't think the difference in production quality is noticeable to the average listener. And that's disconcerting.
Upvote
11
(11
/
0)
Oh wow man I see the future! Middling commercial-style graphics with blandly appealing humans, corporate music and full-throated young-upscale-adult spoken audio, with CGI-fuzzy soporific backgrounds that draw us all toward our golden years.
Last edited:
Upvote
13
(15
/
-2)
What happens when it's easier and less time consuming to create content then it is to consume it? We aren't there yet but it's not far off...quality of the content aside, that's a profound shift of some kind.
Strange times ahead.
Strange times ahead.
Upvote
24
(24
/
0)
Feels a bit like when "transitions" were a novelty and software sold on the basis of how many silly and useless advanced effects it offered, making most hobby videos terminally unwatchable.
Upvote
6
(8
/
-2)
I always get downloaded when I bring this up. But soon we will be at the point where you can have a home designed movie. There will be parameters like action, romance, etc. and you can assign them 1 through 10.
Upvote
-1
(3
/
-4)
The average value of anything and everything tends towards zero, and it also will be the de-facto end of any shared experience of content (like the current fragmentation of media, but then times a million). And we will all be so much better off. /s
Upvote
1
(3
/
-2)
At this point, I’m starting to reasonably expect in articles like this a description of what data was used to train the model. After all, one of the most newsworthy bits is whether the training data was procured in a shady or misleading way. Without that context, I’m not sure exactly how to digest the risks of the tool. Probably a good best practice for articles on the sexy new LLMs rushed out since breakfast.
Upvote
2
(5
/
-3)
I'm pondering a future where 50-80% (and maybe that's a conservative estimate) of content is created by AIs. I wonder if there will even be fads by that time. Unless we come up with AI that create actual new content rather than just derivative remixes. Once there's no money in it for humans making original content (other than to be sucked into a training dataset with no remuneration), the AIs will only be able to crib from each other.
Last edited:
Upvote
11
(11
/
0)
If it is Adobe's AI that will generate the sound effects or music for a certain scene, what will its effect on the ownership/ copyright of the videos?
Upvote
5
(5
/
0)
I was wondering this as well. The entire piece would be copyrighted, but the copyright of some of the individual elements would be in doubt.
I would think something like the photo/video where you tell it "change to golden hour" would be no problem, assuming that the author had the copyright or license to use the original photo (with editing permissions). However, someone else who also had a copyright/license to the same photo could possibly snag your AI-edited copy and use it.
Ditto with it selecting B-roll elements out of copyrighted footage. Those elements should still be copyrighted, because the B-roll is copyrighted.
I'd think the music - which seems completely auto-generated - wouldn't be copyrightable, given the previous ruling.
The storyboards/pre-viz are a little harder. It's possible that those won't be copyrightable given that they are seemingly just output from prompts. But I also don't think that's a very big concern for something like storyboarding.
Last edited:
Upvote
1
(1
/
0)
Long before we are THERE, we'll have entered (we're doing that RIGHT NOW) the phase where it has become almost completely opaque WHERE the still-existing human creativity and skill starts and ends. People will both praise human artists for aspects that were almost completely AI-driven, and reject/chastise genuine human input because it is assumed to be "mostly AI generated anyway". It will be interesting to see how much value people will be willing to assign, and on the basis of which considerations exactly.
Upvote
9
(9
/
0)
I think that snippet is showing colour-correction effects being applied to a film clip, rather than the media itself being AI-generated. Same goes for the "brighten face" bit; it's a secondary CC effect with an AI-generated mask for her face. So, no change of copyright.
---
I use 3D, photoshop, premiere and aftereffects for ~half my job. Premiere's auto-transcription has already been a big help. There are so many tools and disciplines that it's hard to become an expert in all of them. Lowering the barrier for time-intensive and specialist skills, particularly
storyboard and previz, is fantastic. If the latter would be editable (a real-time 3D world) and you could experiment with different camera moves and framing....wow.
Upvote
5
(5
/
0)
Then just make it so the suits don't want your stuff. Let the forces of Rules 34 and 35 thrive in the new age of AI!
Upvote
3
(3
/
0)
That's not "current USA law," says the U.S.-licensed intellectual property attorney, and no amount of all-caps commenting will change that previously reported state of affairs.
The human authored bits still get copyright protection, and often they're so comingled with the generative machine learning model authored materials ("Stable Diffusion"), or the machine learning driven filters and modifications (most Adobe tools), that you can't separate one from the other so as to use only the unprotected, "AI"-generated bits.
Last edited:
Upvote
5
(5
/
0)
When I read this, I heard it in the voice of Moist Critikal
Upvote
2
(2
/
0)
The episodes of 90s-era Star Trek where they walked into an empty Holodeck room and talked through what something should look like, with the computer just instantly creating it and then refining it based on their spoken prompts? Those were some of my favorite moments - just because it seemed so absurdly like magic, even more than warp drives and cloaking devices.
Aside from the whole light<->matter conversion thing, we're getting really damn close to having developed the core technologies that will someday enable that to be real.
Aside from the whole light<->matter conversion thing, we're getting really damn close to having developed the core technologies that will someday enable that to be real.
Upvote
8
(10
/
-2)
Yeah, I predict it won't be very long before we see someone make an AI driven video recreation of this.
Upvote
5
(5
/
0)
The music strikes me as something where you might currently use a library track (even back in the 60s, those albums were indexed by keywords).
But I doubt it will threaten the current market for sync rights or OST composition, as they seem to be thriving - even a mid-market videogame or TV series will commission an OST, because the cost of OST production has come right down.
(It was more common to hear an obviously all-synthetic soundtrack to a program in the 80s - 'keyboards are replacing orchestras'. These days, soundtracks tend to be hybrid, and I guess most composers have already created a guide version in a DAW before spending a small amount of time in a studio with session musicians).
Upvote
2
(2
/
0)
That would require Humans to be creating those fads in the first place in a large enough quantity for the AI to pick up on. If there are no creatives as it is cheaper to employ an AI then there will be no source material for the AI to learn... Not only that but it is entirely possible for rogue states to spike the internet with material that the AI then spreads for them automatically and rapidly. Imagine AI inserting Russian propaganda in your new favourite auto generated Comedy series....
Upvote
-2
(1
/
-3)
- Add bookmark
- Featured
- #39
I'm in Firefly beta, and the AI is, well, very dumb. Some serious hand-holding is required to get it to make what you need, and sometimes it's not following instructions, no matter how detailed. I've spent like 30 minutes trying unsuccessfully to generate a photo of Lord Voldermort wearing a business suit sitting behind a desk in front of a bunch of microphones giving a soccer press-conference. The AI couldn't even produce a simple photo of a bald man that doesn't have a nose - let alone any of the other stuff. I just gave up.
And don't even get me started on prompt restrictions - you wanna see a giant robot blowing shit up with plasma cannons? Well, you CAN'T, it goes against Firefly guidelines ("cannon", "gun", etc.). You want to generate an illustration of a fit girl in gym outfit doing exercises (something I'd actually use a LOT for work, because taking footage at the gym is a pain in the ass to arrange)? Well, you CAN'T, - something in that prompt triggers a block and the guidelines warning (I actually seriously suspect it might be the word "fit" - because judging by what Firefly DID generate, it doesn't like fit-looking women, it's all about body positivity).
I'd really like to see Adobe's AI in question to actually work on/generate video footage THAT efficiently (yes, I realize it's an advertisement) - but there's NO WAY it's going to work as well as shown.
And don't even get me started on prompt restrictions - you wanna see a giant robot blowing shit up with plasma cannons? Well, you CAN'T, it goes against Firefly guidelines ("cannon", "gun", etc.). You want to generate an illustration of a fit girl in gym outfit doing exercises (something I'd actually use a LOT for work, because taking footage at the gym is a pain in the ass to arrange)? Well, you CAN'T, - something in that prompt triggers a block and the guidelines warning (I actually seriously suspect it might be the word "fit" - because judging by what Firefly DID generate, it doesn't like fit-looking women, it's all about body positivity).
I'd really like to see Adobe's AI in question to actually work on/generate video footage THAT efficiently (yes, I realize it's an advertisement) - but there's NO WAY it's going to work as well as shown.
Upvote
21
(21
/
0)
I'm a motion graphics designer, so I'm genuinely curious about this. What happens when I generate the... checks notes... "3D molten steel text" that was requested of me for v1. Then the whopping cohort of 12 reviewers (yes, this is what I currently have on an actual project) all feel like they have to appear like they are "doing work" so they all have to chime in and send back some vague/conflicting notes. Can you make the molten steel slightly less melty? Can you char the bottom edge of the molten steel? Can you make the "o" in molten more melted than the other letters?... etc. Then what do I do? Does this generative process give me parameters that I can manually adjust? Or am I stuck trying to conceptualize more text-based ways of describing the half-baked ideas the reviewers send my way, and sitting there with a sinking feeling at all the time it's taking to generate each version? Or will it just save time to skip the AI and just manually create the effect?
Upvote
6
(6
/
0)