This week on Denoised, hosts Joey Daoud and Addy Ghani shift to a "This Week in AI" roundup format, examining the rapid evolution of AI video generation tools and image models. The episode covers Midjourney's entrance into the video generation space, improvements in video model capabilities from companies around the globe, and discussions about the challenges of maintaining consistency across different AI tools. Additionally, they address ongoing IP concerns as studios take legal action against AI companies, highlighting the tension between innovation and copyright protection in this rapidly advancing field.
Midjourney Video Enters a Crowded Market as IP Lawsuits Loom
Midjourney, one of the original pioneers in AI image generation, has officially launched its V1 video model. This release has generated significant attention in the AI creation community, offering several notable features:
480p resolution output (lower than competitors)
Five-second clip generation with extension capability up to 22 seconds
Two motion settings: high motion and low motion for different effects
Built-in extend feature for longer sequences
The timing is interesting as Midjourney is currently facing a lawsuit from Disney and Universal regarding AI-generated outputs of protected characters like Yoda and various comic book characters. Joey and Addy note this is the first time major studios have actively pursued legal action against AI image generators for their outputs rather than their training methods.
"This is the first time the studios are actively going after them," Addy explains, highlighting that while these lawsuits target specific outputs of protected characters, they don't address the fundamental issue that the models have already been trained.
Joey suggests the focus of regulation will likely settle on controlling outputs rather than halting model development: "I still think the focus is on the outputs... it's like, sure, put blockers in these things where you can't generate IP-protected characters."
Key takeaways:
Midjourney's video model is more limited than competitors at launch but represents a major player entering the video generation space
Disney/Universal lawsuits against Midjourney may set precedents for how IP will be handled in AI generation
Studios must balance IP protection with embracing AI innovation that could benefit their industries
Global competition makes complete regulation difficult as Chinese models face fewer IP restrictions
Video AI Models Advance with Better Physics and Resolution
The podcast highlights several significant updates in the video AI model space, showcasing how quickly the technology is improving:
This Chinese company's second-generation video model boasts several improvements that address common weaknesses in AI video generation:
1080p output resolution (higher than Midjourney's 480p)
Improved instruction following capability
Enhanced physics mastery for realistic movement
Demo clips showing complex physical movements like gymnastics and circus performances
"Extreme physics mastery," Joey notes, was intentionally highlighted in their demos because "gymnastics was like one of the weak spots with a lot of these models."
TikTok's parent company has quietly released a model that's earning high rankings in technical evaluations:
Top ratings in artificial analysis for text-to-video and image-to-video
Multi-shot views capability, allowing different angles of the same scene
High-motion capabilities, including realistic surfing scenes
Diverse style options
Benchmarking Challenges
The hosts discuss the need for standardized benchmarks to evaluate these rapidly evolving models. Common test cases they use include:
Knife cutting vegetables
People eating spaghetti
Pressing buttons
Water movement and physics
Gymnastic movements
Joey notes: "I guess it's useful to have a couple of baseline [prompts]," while Addy suggests testing for specific technical capabilities: "If you generate a sunrise in a dark environment... that would be a good indicator of dynamic range."
Key takeaways:
Video AI models are rapidly improving on previously challenging physics simulations
Chinese companies are producing competitive or superior models to Western counterparts
Higher resolution outputs (1080p) are becoming standard in newer models
The industry lacks standardized benchmarks for fair model comparison
Topaz, Flux, and Krea: Tools for Better AI Output Quality
Beyond new video models, the hosts discuss several tools designed to enhance or improve AI-generated content:
This new cloud-based offering from Topaz Labs builds on their reputation for AI upscaling:
Features a creative slider that allows users to balance between faithful upscaling and creative interpretation
Cloud-based processing with credit system (versus their desktop software which runs locally)
Particularly useful for enhancing lower-resolution AI video outputs
Has already been used to help AI-generated content pass broadcast quality control standards
Addy explains the likely technical approach: "The upscale is still happening and then it's weighing what the creative upscaling is and then mixing between the two."
Flux's new image model introduces a significant capability that changes how consistency works in AI generation:
Eliminates the need for separate fine-tuning through lightweight LORA models
Maintains contextual awareness between prompts
Allows users to reference previous generations in new prompts
Requires cloud processing rather than running locally
"You generate, let's say, cartoon Joey in Flux Kontext, and you can call that cartoon Joey. And then on the next prompt say cartoon Joey is now going to a bar. And it remembers it, it has contextual awareness," Addy explains.
The company known for real-time generation now has its own proprietary image model:
Focuses on reducing the "AI look" in generated images
Produces more realistic skin textures and imperfections
Handles lighting and specular highlights more naturally
Competes with Midjourney for stylistic range but with more photorealistic options
"Krea is taking into account specular highlights and direction of lighting really, really well. Skin blemishes," notes Addy, explaining why their outputs often look more natural than competitors.
Key takeaways:
Cloud-based processing is becoming the norm for advanced AI generation tools
Models are increasingly incorporating ways to maintain consistency between generations
The industry is addressing the "AI look" by focusing on realistic imperfections and lighting
Workflow integration across different tools remains challenging for professional users
Consistency Challenges for Professional AI Users
A recurring theme throughout the episode is the difficulty of maintaining consistency when working across multiple AI tools. Joey shares his experience trying to create professional content:
"Sometimes I will go to a bunch of different tools to get the output that I'm looking for. But then the problem is like, they all have slightly different qualities and then it just feels a little less consistent in the output... if I'm combining shots from Luma and combining shots from Veo, there's not a consistency across the whole sequence."
This challenge mirrors traditional filmmaking concerns, as Addy points out: "You can't go from a cut from a Blackmagic camera to a RED camera unless you have a really good colorist that can match the two."
However, AI outputs present a unique challenge: "The AI generation is so baked in, it's hard to modify that," unlike traditional footage where more data is available for color grading.
The hosts discuss how professional workflows are still developing, with Joey noting: "If you're doing a production that involves AI, like you're probably gonna wanna do test shoots... figure out what that workflow is gonna be so that we can dial it in and kind of have, be consistent."
Key takeaways:
Different AI models have distinct "looks" similar to different film stocks or camera sensors
Maintaining visual consistency across tools remains a significant challenge for professional users
Established workflows for testing and standardizing outputs are still developing
The lack of "raw" data in AI outputs makes post-processing more challenging than with traditional footage
The Future of AI Characters and Digital Actors
The episode concludes with a discussion of Arcads AI, a platform focused on digital actors, which leads to broader questions about the future of AI-generated personalities:
Who owns the rights to AI-generated characters and personalities?
How will talent agencies adapt to representing virtual talent?
Could AI-generated influencers impact the current influencer economy?
What legal frameworks will emerge around AI character ownership?
Addy questions: "If we create digital talent, synthetic talent, can a talent agency represent them? And then can their likeness be used for product photography or advertisement - who owns the likeness?"
The hosts reference the Kalshi ad that recently aired during NBA games, created entirely with Veo 3, as an example of AI-generated content entering mainstream advertising.
Joey suggests that human authenticity may become more valuable in a world of AI content: "If everything's sort of commoditized and it's easy to spin up a newsletter, it's easy to spin up a YouTube channel... the counter to that would be the human connection or the human person behind it."
Key takeaways:
Legal and business models for AI-generated characters are still developing
Major brands are already using AI-generated content in broadcast advertising
The tension between authentic human content and AI-generated material will define future media landscape
Questions about ownership and representation of AI characters remain unresolved
Conclusion
The rapid evolution of AI video and image generation tools shows no signs of slowing, with new models emerging weekly from companies around the globe. While technical capabilities continue to improve—addressing previous limitations in physics, resolution, and realism—significant challenges remain for professional users trying to maintain consistency across different tools. Legal questions about intellectual property and character ownership add another layer of complexity to this rapidly developing landscape.
For creative professionals navigating this space, staying informed about the distinct capabilities of each tool while developing standardized workflows will be essential for successful implementation. As Joey and Addy demonstrate throughout this episode, understanding both the technical capabilities and practical limitations of these tools is crucial for anyone working at the intersection of AI and media production.
The views and opinions expressed in this podcast are the personal views of the hosts and do not necessarily reflect the views or positions of their respective employers or organizations. This show is independently produced by VP Land without the use of any outside company resources, confidential information, or affiliations.
