Hey AI video people. Who can make this?

[newsmike]

These videos have been all over social media. Who can make one of these? 

 

[coder_nazim]

Mind-blowing performance😄

[rawque_gulia]

2 hours ago, newsmike said:

These videos have been all over social media. Who can make one of these? 

"Viggle AI" can do this. For now, it only works through Discord but their dedicated app is launching soon. I have created a few videos for fun and it's amazing haha!

[uk1000]

Whatever they used they probably used an existing dance video that the AI used to reference and create one similar to it. Maybe animatediff can do something like that.

Here's a few very short clips that are a bit similar (from stable video diffusion, which isn't very accurate and can only create really short clips).

It's mostly a robot trump.

Edited September 7 by uk1000

[newsmike]

On 9/7/2024 at 10:34 AM, uk1000 said:

Whatever they used they probably used an existing dance video that the AI used to reference and create one similar to it.

 

Correct. This is the source video:

 

 

[Kesha]

Hi @newsmike! Unfortunately, these kinds of AI videos are against our Terms of Service. 

[newsmike]

7 minutes ago, Kesha said:

Hi @newsmike! Unfortunately, these kinds of AI videos are against our Terms of Service. 

I'm not sure I understand, can you please explain so I may understand how and why?

[uk1000]

37 minutes ago, Kesha said:

Unfortunately, these kinds of AI videos are against our Terms of Service.

Isn't parody a copyright defence though (perplexity says it's a recent exception to copyright law in the UK)? And perplexity.ai says:

Quote

Parody has long been recognized as a potential "fair use" defence against copyright infringement claims in the US

Like doesn't Fiverr allow certain content if it's parody that it otherwise wouldn't (due to copyright)? This original video was one that the person in it (one of them) posted to his twitter I think, showing he didn't mind it.

The community standards says about AI:

https://www.fiverr.com/community/standards/ai-generated-content

Quote

Both sellers and buyers are prohibited from using AI tools to spread misinformation, nonconsensual content, create deepfakes (in a form of video, photo, or audio recording), fake accounts, or generate faces of humans who do not exist for unlawful purposes.

Is that what it's against? But it's not spreading misinformation and if it's parody it should be okay copyright/IP-rights wise (especially if the person in it is okay with that type of content). And it's not for unlawful purposes if it's parody surely.

This is what perplexity.ai says about a video like that with UK copyright law:

Quote

UK Parody Exception
The UK has a specific copyright exception for parody, which was introduced in 2014. For a work to qualify under this exception, it must meet certain criteria:

    It must evoke an existing work while being noticeably different from it.
    It must constitute an expression of humor or mockery.
    It must be considered "fair dealing"

Application to the AI Video Scenario
Given these criteria, an AI-generated video of Donald Trump doing a silly dance in the street could potentially qualify for the parody exception in the UK:

    Evoking existing work: The video would likely evoke Trump's public persona and typical behavior, while being noticeably different by depicting him in an unusual, comedic situation.
    Humor or mockery: A silly dance video would almost certainly be intended as humorous and could be seen as mocking Trump's public image.
    Fair dealing: This aspect would depend on various factors, such as the extent of copying from original sources and whether the parody competes commercially with any original work

Additional Considerations

    Transformative nature: The parody exception requires the work to be sufficiently transformative and not simply reproduce substantial parts of the original

An AI-generated dance video would likely meet this requirement.
Commercial competition: UK courts would consider whether the parody competes commercially with any original work it's based on

    . As long as the video isn't monetized in a way that directly competes with Trump's own content or likeness rights, this shouldn't be an issue.
    AI-specific concerns: The use of AI to generate the video adds complexity, as it raises questions about copyright in AI-generated works. However, the parody exception itself doesn't specifically exclude AI-generated content.

Given these factors, an AI-generated video of Trump doing a silly dance would likely have a strong case for qualifying under the UK's parody exception, especially if it's clearly intended as humorous commentary on his public persona and doesn't compete commercially with any original works.

Edited Wednesday at 05:33 PM by uk1000

[newsmike]

I believe usage matters. If I wanted to post it publicly, yes 100%. But if I wanted to do this to show to a friend for a few laughs at their birthday party, there is no copyright issue. 

Edited Wednesday at 05:26 PM by newsmike

[imagination7413]

Just a guess here, but I'd suspect the part Kesha is refering to is "Misleading or Deceptive Services - Deepfakes"

https://www.fiverr.com/community/standards/prohibited-services 

(Though, that's guidelines for gig services, not meme sharing.)

((Unless it's referring to a specific grabby 'dance move' that's kinda bordering inappropriate.))

[uk1000]

So if it was just a funny dance with a famous person like who was in the video that was posted on twitter, and if it didn't have anything inappropriate, and it was obviously not the real person and just meant to be funny, then I think it should be okay (because of the parody exception to copyright law).

If it was an intentional AI video or deepfake trying to look as real as possible, trying to mislead people (eg. promoting something that the person didn't promote or fake news) then that would definitely be against Fiverr rules.

[Kesha]

21 hours ago, newsmike said:

I'm not sure I understand, can you please explain so I may understand how and why?

Sure! I'd be happy to clarify. As others have mentioned, a video like this is considered a deepfake, and we do not currently allow that service on our platform. I understand that these videos are often used for harmless entertainment purposes, but unfortunately, it's difficult to regulate the intent and ethics behind them. For that reason, we don't allow them all together.

[uk1000]

1 hour ago, Kesha said:

As others have mentioned, a video like this is considered a deepfake, and we do not currently allow that service on our platform

Isn't Fiverr's definition of a "deepfake" more than what one really is?

This: https://www.fiverr.com/community/standards/prohibited-services

says:

Quote

Deepfakes

Any video, photo, or audio recording that has been substantially modified by altering its sequence or composition, or by modifying the original audio in a way that fundamentally changes the original media’s understanding, meaning, or context (including those created with the help of generative AI tools and programs).

Due to the high risk of spreading misinformation, non-consensual content, and illegal activity through deepfake techniques, Fiverr does not allow any creation of deepfakes on the platform.

it calls a deepfake "Any video, photo, or audio recording that has been substantially modified by altering its sequence or composition...in a way that fundamentally changes the original media’s understanding, meaning, or context".

yet Fiverr has whole categories for modifying the sequence of video clips or video compositing, photo manipulation.

So what Fiverr says isn't allowed is "Any video, photo, or audio recording that has been substantially modified by altering its sequence or composition...in a way that fundamentally changes the original media’s understanding, meaning, or context" but it has a subcategory for video compositing https://www.fiverr.com/categories/video-animation/visual-effects/video-compositing?source=category_filters

where Fiverr says:

Quote

Combine multiple visual elements and blend them into a stunning piece

which can and often will be substantially altering the original composition to make things that weren't there before but in a legitimate way for fiction etc. Look at any design gig's gig image and they'll likely have many manipulated images (eg. flyer design, with images not specifically created to be put on flyers). What Fiverr seems to say isn't allowed above. Adding an effect that a buyer asked for to a shot video would be altering the original meaning of the source clip (adding something that wasn't there originally), which seems like it's against that rule above, but it's what Fiverr has a whole subcategory for.

Fiver also has photo manipulation sbucategory where it says:

Quote

Photo Manipulation

From filter & effects to compositing, our experts will twist and tweak any photo to suit your needs

yet Fiverr's rules say it's not allowed to alter the "composition...in a way that fundamentally changes the original media’s understanding, meaning, or context". But that is what people will be asking for in the photo manipulation - like totally changing the backgrounds of photos etc.

Surely Fiverr should change the rules to only disallow it if it's presenting that altered video/photo as fact/news not if it's just meant to be fiction/a meme/funny video/artistic composition.

Edited Thursday at 03:25 PM by uk1000

[uk1000]

In Fiverr's own webinar (Joinining Forces with Geneative AI): https://help.fiverr.com/hc/en-us/articles/13235854504209-Joining-Forces-With-Generative-AI

at 40:15 it shows one of the people taking original normal photos that they had of themselves in different settings and then using those to create AI images in a brand new settings automatically, and an AI video based on them.

but Fiverr's current rule about AI "deepfakes" says:

Quote

Any video, photo, or audio recording that has been substantially modified by altering its sequence or composition, or by modifying the original audio in a way that fundamentally changes the original media’s understanding, meaning, or context (including those created with the help of generative AI tools and programs)....Due to the high risk of spreading misinformation, non-consensual content, and illegal activity through deepfake techniques, Fiverr does not allow any creation of deepfakes on the platform.

eg. in the webinar linked above they say (using Whisper to audio to text):

Quote

One of the fun things about Astria is that you
can train it on your own images. So I took Astria, gave it
basically those images you see here, a bunch of pictures of me
and it generated this model that can generate pretty realistic
images of me in different styles [ie. putting himself it totally new scenes, totally changing the original intent of the source images]

And you can even prompt it basically like with your own text. For example, if I want a
close up of my face... it will take it a few
seconds to generate. It's pretty easy to train it. You can see
here, you do a new fine tune, describe it, upload the images,
and it costs $1.50 for the training and $0.10 per prompt,
which gives you eight images. It can also generate video and
when video basically is looking something like this.

Later in the same video he uses AI to automatically turn his image into a cyborg (changing the original intent of the source image, against what Fiverr's rules say we are allowed to do), saying (at 44:18 approx):

Quote

It can generate images from the same way that Astria can.
However, it has one additional algorithm called Instruct
Pix2Pix, which can give an image, give it instructions of
how to edit. For example, turn him into a cyborg.
And about 10 seconds and I am a cyborg.

Then in the webinar he shows how to totally a person from a video using AI (RunwayML) (changing the original intent and composition of the source video from what it was originally, despite Fiverr's rule about changing content "in a way that fundamentally changes the original media’s understanding, meaning, or context (including those created with the help of generative AI tools and programs)".

Edited Thursday at 04:45 PM by uk1000

[Kesha]

Hi @uk1000 Thanks for bringing up those points! While some photo manipulation is allowed on the platform, this kind of video falls into artificial footage, and not manipulated media. Thus, our reasoning for not allowing this on our platform stems from the goal to keep misinformation and deep fake content from spreading. We understand the standpoint that some can be used fictionally, however, Fiverr is not able to ensure how the videos are used post-platform, so our general rules do have to be more rigid. Thanks for your understanding.

1 hour ago, uk1000 said:

In Fiverr's own webinar (Joinining Forces with Geneative AI): https://help.fiverr.com/hc/en-us/articles/13235854504209-Joining-Forces-With-Generative-AI

at 40:15 it shows one of the people taking original normal photos that they had of themselves in different settings and then using those to create AI images in a brand new settings automatically, and an AI video based on them.

but Fiverr's current rule about AI "deepfakes" says:

eg. in the webinar linked above they say (using Whisper to audio to text):

Later in the same video he uses AI to automatically turn his image into a cyborg (changing the original intent of the source image, against what Fiverr's rules say we are allowed to do), saying (at 44:18 approx):

Then in the webinar he shows how to totally a person from a video using AI (RunwayML) (changing the original intent and composition of the source video from what it was originally, despite Fiverr's rule about changing content "in a way that fundamentally changes the original media’s understanding, meaning, or context (including those created with the help of generative AI tools and programs)".

 

[uk1000]

30 minutes ago, Kesha said:

this kind of video falls into artificial footage, and not manipulated media.

So what exactly is allowed and what is disallowed? shouldn't that Fiverr rule be made clearer on that page?

Fiverr states "Any video, photo, or audio recording that has been substantially modified by altering its sequence or composition, or by modifying the original audio in a way that fundamentally changes the original media’s understanding, meaning, or context..."[isn't allowed], yet that's what multiple categories of gigs do.

Fiverr themselves (in that webinar) alter video using AI and alter images of people using AI to turn themselves into a cyborg (why if it's against Fiverr's rules?), and use AI to remove people in a shot where the original composition and intention of that shot had been to have that man in it. If some things are allowed they should change that rule, so all editing gigs and compositing gigs won't be breaking their current rule about "substantially modified by altering its sequence or composition...in a way that fundamentally changes the original media’s understanding, meaning, or context."... not being allowed. The rule implies it isn't just about using AI, but non-AI methods too, yet Fiverr has whole image editing and video compositing categories to allow for substantially altering images/video in different ways, including adding things to scenes that were never originally going to be put into those specific scenes.

30 minutes ago, Kesha said:

We understand the standpoint that some can be used fictionally, however, Fiverr is not able to ensure how the videos are used post-platform, so our general rules do have to be more rigid.

so now I'm not allowed to edit fictional scenes to alter their composition or add elements to them (that weren't in the original scene or originally thought of before a buyer then asks for them) as it would be breaking Fiverr's rule? Are all compositing and image editing categories going to be disallowed because of this rule? Do I need to refuse to add any elements to (or remove any from) scenes now?

since it says:

Quote

(including those created with the help of generative AI tools and programs).

meaning ones edited normally (without AI) would also be part of that rule (unless the rule is changed).

Edited Thursday at 06:05 PM by uk1000

[newsmike]

Wouldn't all of these and hundreds more in the AI section of Video Services be far more "fake News" and disinformation?