Filipe Reis’ Post

I see people on LinkedIn, even professionals with experience in the industry, talking about Sora and panicking without reason, without analyzing and breaking down that AI. In 4 years, the same errors persist, and they always have. Because that's how AI works. It's just copy-paste without intelligence behind it. Just stupid algorithms. Yeah, because the algorithms don't have synapses, experience, empathy, or hard work. They're just algorithms. The same ones that ban you from Facebook for 30 days because they misunderstood a word. As I explained in a previous post, this crap is useless for any kind of production (overall NDA). It's good to show to friends if you're not in the industry. "Wow, look at what we have here in 2024!" But if you're a professional, instead of whining about some apocalyptic future where we'll all be replaced and spreading negativity and anxiety, take the time to break down all the videos of Sora and give feedback on them. This kind of post could create a better status for yourself and others, instead of posting mediocrity and saying "omg we are fucked!!" To put it more trivially, "Put your shit together and give yourself a better attitude." I'll be posting some feedback videos in the next few days.

On top of being unreliable and ethically dodgy, there is another issue in the creative space. The sheer tsunami of AI-generated images is crowding an already super-mega crowded attention economy. In an age where creatives have been pushed into working in the attention economy and in which our livelihoods depend on how well we do on social media platforms (Read the article "Everyone's a Sellout Now" by Rebecca Jennings), with ever diminishing slices of the pie, some of these very platforms we thought safe and specific for artists have made it impossible to build audiences or attention from lack of moderation, and ongoing enshittification. Behance, Instagram, DeviantArt, Twitter, ArtStation are all now much harder to garner any momentum in. For emerging talent and those just joining the industry, the game is ever harder, as the over saturation of AI images will slowly crowd us out.

While the reasoning below holds true to a degree for bigger budgets (for now), Sora and whatever follows, will certainly prove very disruptive for people working with smaller clients on lower budget projects. This is a huge sector of the industry that will be turned on its head. This in turn will result in higher competition for the big budget projects, which means that supply will outstrip demand, resulting in worse outcomes for every artist involved, including the ones that currently feel safe working on bigger budget productions. Let's also consider the fact that AI visuals have evolved rapidly in a space of just a few years, with leaps and bounds in visual quality taking place in just a few recent months. The pace is always accelerating. Sure there are issues with the video below, but these problems are highlighted by an artist, someone who has a trained eye for these things and years of experience. Show this to 90% of the rest of the population - they won't care or notice. And what would stop the Sora prompter to spit out another 1000 variations until it fits the vision of the client? Everyone should draw their own conclusions of course, but I think it'd be wise to at least start getting a little bit worried.

Let’s just start with this. I am wholly unimpressed with Sora. And I have to agree with the post below, having done video editing and game cinematics, the notion that an AI tool is affecting any professional work, at this stage or in the next few years is patently ridiculous. The people saying this are people who don’t know how to make videos at all. To them it’s a dark art. And anything that makes anything is exciting to them. Largely non-iterative, clumsy amateur work in video form will mostly be useful for the lowest common denominator of online advertisers in social media, and to some extent amateur content creators. Internet marketers that do the lowest common denominator of animation need to be worried about Sora. Nobody else does. It’s not going to make a feature film. It’s not going to make anything that will draw money. It’s not going to be used in games, animation, film, or TV, anywhere, professionally, except by amateurs that are trying to vaguely look like professionals. No one is going to buy a ticket to contribute to box office sales using anything from these tools. But it will appear in spam. It will appear in cheap online advertisements. Everywhere. It will be used by marketing from brands that have no budgets. It will be used by amateurs everywhere to create a deluge of video crap that you will wish you didn’t have to see. That’s the place for AI video. Cheap entries into social media content and advertising. Generally a space with low standards or no standards. Places with no budgets. Great. Like we really need more of that. AI tools will contribute to a flood of cheap crap and shitty that you will be trying to avoid. Let’s step back and look at this. People who do not do creative endeavors are impressed with anything that looks vaguely creative. The thing is that creative work should have iteration behind it. And that iteration is necessary to get it into the right final state for a client or audience. Iteration is intentional. There is no iteration on that level with AI tools.

I recently shared my thoughts with Guillaume L. at millefeuille.ai. We talked about the use of AI in art, Animaj innovative tools for 3D character animation and the future developments we have in store. You can read the full interview here (available in French): https://lnkd.in/es9_aRqW #3DAnimation #AI

AI FROM PICTURE TO VIDEO

🖍️✨ Bringing Childhood Imagination to Life! 🎨🎥 Just when I thought my daughter's drawings couldn't get any more magical, I stumbled upon Meta AI Research's Animated Drawings. 🌈👧 Picture this: her latest masterpiece, a vibrant drawing of a quirky, lopsided cat, suddenly prancing around in a digital animation! 🐱💃 It's like watching her imagination leap off the paper and dance across the screen. Using this demo, based on advanced algorithms like Detectron2 and AlphaPose, I transformed her static artwork into a lively animation. The process? As simple as uploading the drawing, and voila - an animated adventure begins! It's a surreal experience seeing her crayon creations come to life, moving with a charm only a child's drawing could possess. It's like opening a door to a whimsical world where her doodles are the stars of their own animated show. And the best part? I get to keep the privacy intact while sharing this joyous creation. I'm already excited to see what other delightful animations we can create together. Maybe next time, her drawing of a three-eyed alien will take the stage! 👾🌟 URL: https://lnkd.in/dbammSVU #AnimatedDrawings #MetaAI #ChildhoodCreativity #DigitalMagic #FamilyFunTime

Here is an interesting article by Valentin ValDo Dornel for SUPAMONKS which gives us an overview of the use of Unreal Engine in studios and of AI applied to 3D animation.

Media2Face: Co-speech Facial Animation Generation with Multi-Modality Guidance The #Media2Face project advances 3D facial #animation using #multimodal data, overcoming data scarcity and blending challenges to enhance realism and flexibility. Their method involves a system called Generalized Neural Parametric Facial Asset (GNPFA), a sophisticated tool designed to map facial features and movements to a universal expression space. This means it can differentiate between who you are (identity) and how you express yourself (emotion), a crucial step for creating diverse and realistic #avatars. Secondly, by applying #GNPFA across a vast collection of #video data, the researchers were able to harvest a rich dataset named M2F-D. This dataset stands out for its variety and the depth of its emotional and stylistic annotations, laying the groundwork for more nuanced animations. The M2F-D dataset, derived from extensive video analysis, underpins nuanced animation development. While promising high-fidelity animations adaptable to various styles, challenges like capturing the subtleties of facial expressions and maintaining believable lip-sync highlight the ongoing struggle for lifelike realism. This approach also confronts the "uncanny valley" through its parametric model, necessitating manual adjustments to achieve authentic asymmetry and individuality. Continuous innovation and a clever data handling strategy are essential for progressing towards more realistic real-time animation and lip-sync. (The reasons why the #parametric approach has issues is beyond the scope of this post, because it would become longer than could fit here.) Be sure to check out the website for a lot of other static and video samples. I think one of the key things that stands out, is that this approach can yield extreme #blendshapes that are unnatural, and/or do not properly coincide with what is being said and how. Yet, with more examples, more annotations, etc., compression techniques, etc., it’s worth watching. Note that code and dataset for this are supposed to be upcoming, but it does not yet state when or under what license that might be. Website: https://lnkd.in/gkWtPcDN #Research Paper: https://lnkd.in/gxjFEw3Y _ #digitalhuman #digitalhumans #facialanimation #3dmodel #3danimation #multimodal #mocap

Media2Face: Co-speech Facial Animation Generation with Multi-Modality Guidance The #Media2Face project advances 3D facial #animation using #multimodal data, overcoming data scarcity and blending challenges to enhance realism and flexibility. Their method involves a system called Generalized Neural Parametric Facial Asset (GNPFA), a sophisticated tool designed to map facial features and movements to a universal expression space. This means it can differentiate between who you are (identity) and how you express yourself (emotion), a crucial step for creating diverse and realistic #avatars. Secondly, by applying #GNPFA across a vast collection of #video data, the researchers were able to harvest a rich dataset named M2F-D. This dataset stands out for its variety and the depth of its emotional and stylistic annotations, laying the groundwork for more nuanced animations. The M2F-D dataset, derived from extensive video analysis, underpins nuanced animation development. While promising high-fidelity animations adaptable to various styles, challenges like capturing the subtleties of facial expressions and maintaining believable lip-sync highlight the ongoing struggle for lifelike realism. This approach also confronts the "uncanny valley" through its parametric model, necessitating manual adjustments to achieve authentic asymmetry and individuality. Continuous innovation and a clever data handling strategy are essential for progressing towards more realistic real-time animation and lip-sync. (The reasons why the #parametric approach has issues is beyond the scope of this post, because it would become longer than could fit here.) Be sure to check out the website for a lot of other static and video samples. I think one of the key things that stands out, is that this approach can yield extreme #blendshapes that are unnatural, and/or do not properly coincide with what is being said and how. Yet, with more examples, more annotations, etc., compression techniques, etc., it’s worth watching. Note that code and dataset for this are supposed to be upcoming, but it does not yet state when or under what license that might be. Website: https://lnkd.in/guPtbb6R #Research Paper: https://lnkd.in/gEpv38VK _ #digitalhuman #digitalhumans #facialanimation #3dmodel #3danimation #multimodal #mocap