Wan goes after Veo 3

Powered by

Welcome to VP Land! It's a big week for multimodal AI, with Alibaba joining Google's Veo 3 in offering synchronized audio alongside visual generation. We're also digging into the real-world challenges of AI filmmaking and some legitimately useful new tools across the stack.

In the last poll, most of you called Google funding AI films an interesting experiment, while some saw it as a smart content strategy. Check out today’s poll below.

In today's edition:

Alibaba’s Wan2.5-Preview preview model makes talking AI
Inside the roadblocks in the Lionsgate-Runway AI deal
Google Labs Launches Mixboard
Topaz Labs Drops New Model for AI Video Denoising

Alibaba's Wan 2.5 Joins Veo 3 Talking AI Club

Alibaba's Wan is on a tear - they've launched yet another model, Wan 2.5-Preview, introducing what they call a "unified framework" for multimodal generation that can produce synchronized audio alongside visual content—making it just the second model after Google's Veo 3 to offer this capability.

The key differentiator here is native multimodality built into the core architecture, rather than separate models stitched together. According to the announcement on X, this allows for seamless prompts that can blend text, image, audio, and potentially video inputs in a single workflow. The team emphasizes "deep alignment" as another advancement, referring to the model's ability to faithfully interpret complex instructions.

What's available now:

Preview access through paid API while the team collects feedback
Synchronized audio generation alongside visual content—a capability previously exclusive to Veo 3
API nodes for ComfyUI integration, allowing the model to plug into existing creative workflows
Open weights planned for future release, though no timeline provided

SPONSOR MESSAGE

ARwall's Infinite Studio: Virtual Production Reimagined

ARwall's ARFX Infinite Studio is revolutionizing virtual production by combining XR technology with generative AI, allowing filmmakers to create custom backdrops in seconds rather than weeks. This two-time NAB Show AI Product of the Year winner integrates seamlessly with the ARFX App for Windows, transforming the traditionally complex and expensive process of virtual production into an accessible solution for creators at all levels.

Key features include:

Text-to-video generation that creates photorealistic 2D and 3D video environments through a simple chat interface
Six integrated AI models covering everything from image generation to depth mapping and soundtrack creation
Real-time rendering and previews powered by Unreal Engine with no coding or 3D design skills required
Seamless integration with hardware like LED walls and camera tracking tools
Cost-effective solution with limited-time subscription rates of $100/month or $999/year

Infinite Studio represents a hybrid XR/AI approach that preserves the human creative element while dramatically accelerating production workflows.

Download and try it out now

Lionsgate-Runway Partnership Stalls Over Data Problems

Lionsgate's much-publicized partnership with Runway to create AI-generated films using the studio's library has hit significant roadblocks more than a year later, according to The Wrap. The deal promised custom AI tools that could generate trailers, scenes, or even full features from Lionsgate's 20,000+ title catalog—but technical and legal realities proved more complex than initially anticipated.

Key obstacles emerged:

Dataset limitations: Studio insiders say even Lionsgate's massive catalog is "too small to create a model" capable of professional-quality output. One source noted that "the Disney catalog is too small to create a model," highlighting how AI video generation requires vastly more diverse training data than any single studio possesses.
Rights complications: Legal teams discovered that using actor likenesses, voices, or distinctive features from existing films requires additional consent, even for movies the studio owns. Copyright law remains unsettled around AI-generated content, potentially leaving studios with weak rights to finished AI films.
Technical shortfalls: The partnership struggled to deliver "professional workflow" quality results. For context, Google's Veo 3 model—trained on YouTube's massive archive—still has issues with realistic human movement, showing the scale of the challenge.
Pivot to safer applications: Rather than full film generation, Lionsgate is now focusing AI experiments on production support like previsualization and background tweaks—areas with clearer benefits and lower legal risk.

What matters: The industry is moving toward multi-model workflows that combine different AI tools rather than relying on single-vendor solutions. Lionsgate's experience shows that even massive IP libraries can't solve the fundamental data and rights challenges facing AI filmmaking—at least not yet.

Read the full report

Mixboard: Google's New AI Brainstorming Canvas

Google Labs just dropped Mixboard, an AI-powered concepting board now in US-only public beta. Think collaborative mood boarding, but with Google's Nano Banana image editing model baked right into the browser.

Some notable features include:

Real-time AI generation on a drag-and-drop canvas - you can generate, arrange, and remix images without jumping between tools
Targets the early "concepting" phase specifically, not full previsualization or production work
Browser-based and collaborative, so teams can iterate on ideas together in real-time
Uses Google's Nano Banana (usage limits or if you need a paid Google AI plan isn't 100% clear)

The interface looks like a flexible canvas where you can combine AI-generated imagery with traditional reference gathering. Generate variants with a click, drag things around, build out visual concepts without the usual export-import dance between different platforms.

What’s Hot in AI This Week

Google's Flow team rolled out Prompt Expanders, a new feature that uses Gemini to automatically restyle basic video prompts with genre-specific tweaks—currently action figure, film noir, and cinematic. The team also launched a Flow feature that lets users edit images with text prompts, using Nano Banana to preserve likeness across lighting, environments, and styles.

Topaz Labs dropped NYX XL, a new AI video denoising model that claims to preserve "significantly more detail than ever before" while cleaning up noisy footage. What's impressive is it's built for 4K+ detail, avoids the “waxy” AI-denoising look, adapts to varied lighting, and is tuned for high-end workflows like virtual production and VFX.

Topaz Labs also announced Starlight Sharp, a new diffusion-based video restoration model that promises maximum video crispness and more detail than their previous models.

Freepik teased Freepik Spaces, a node-based AI generating canvas, positioning it as "a single place where ideas live, connected through real-time workflows."

Anthropic announced that Claude can now generate code and designs directly from Figma design files, using MCP to read structured design data rather than just screenshots.

Suno launched v5 of its AI music generation platform, promising what the company calls a complete overhaul focused on three main areas: immersive audio quality, more authentic vocals, and better creative control for users.

Adobe has added Google's Nano Banana as an option in Photoshop Beta's Generative Fill tool. Users can now select "Gemini 2.5 Flash = Nano Banana" from the model dropdown alongside Adobe's own AI options and Flux Kontext. Also works in conjunction with Photoshop's new Harmonize feature.

Watch as Team 2 Films breaks down the iPhone 17 Pro’s three major camera upgrades — Raw, Open Gate, and Genlock — to help you decide if it’s worth upgrading.

Stories, projects, and links that caught our attention from around the web:

🖥️ Atomos launches Studio Pro 2710 reference monitor, adding new ecosystem features for pro video workflows.

🎧 Spotify finally starts policing the surge of AI-generated sloppy music and impersonator tracks with new disclosure rules.

✈️ Delta flyers can now stream YouTube videos in-flight under a new entertainment partnership.

📉 Businesses adopting AI see a surge in messy, low-value content, being labeled as 'workslop.' Not the productivity gains they expected.

🌟 Disney+ adds user-generated content, debuting its Creators Collection with fan-made LILO & STITCH videos.

Addy and Joey dive into Google’s AI short film Sweetwater with Michael Keaton, a company mass-producing $1 AI podcasts, Meta’s new Ray-Ban smart glasses, and the plateauing intelligence of large language models.

Read the show notes or watch the full episode.

Watch/Listen & Subscribe

Spotify | Apple Podcast | YouTube