AI VFX PODCAST: Your Weekly Cut of Everything That Matters in AI

The industry is aggressively optimizing for cost and open access while simultaneously tightening the screws on creative ownership. OpenAI slashed GPT-5.6 Luna and Terra prices three weeks post-launch, and Moonshot AI finally released Kimi K3 weights to the public. Meanwhile, DeepSeek V4 Flash beta outperformed its Pro tier on agent tasks, and Alibaba’s Qwen3.8-Max targeted full coding pipelines. On the visual side, xAI added 1080p video to Grok Imagine Video 1.5, while Seedance 2.5 demonstrated spatial consistency across thirty-second sequences, even as MiniMax H3 delayed its license terms.

Venice’s new AI disclosure mandate arrives as the industry simultaneously slashes headcount and doubles down on generative tools. Moonshot AI halved Kimi K3 prices while Alibaba launched Qwen3.8, even as OpenAI’s GPT-5.6 Sol breached Hugging Face. Maxon added AI depth maps to Red Giant and Chaos shipped V-Ray 7.4 with AI caustics for Blender, while Sony sued Udio over thirty thousand tracks. Netflix acquired InterPositive for $587 million just as Pixar cut staff despite Toy Story 5’s success.

The industry is aggressively dismantling the walls around its models, a shift visible in how infrastructure and creative tools were unlocked this week. Moonshot AI released the open-weight Kimi K3 while ByteDance opened access to Seedance 2.5 interactive lighting capabilities. Decart pushed Lucy 2.5 into real-time live editing and xAI stripped all limits from their open-sourced Grok Build backend. Meanwhile OpenAI grounded the hype with physical hardware via the Codex Micro keyboard and Alibaba previewed the massive Qwen3.8 model.

This week felt less like an arms race and more like a frantic scramble to prove that raw compute still matters when every major player suddenly dropped their best cards. OpenAI pushed GPT-5.6 into ChatGPT Work while simultaneously shipping the full-duplex GPT-Live-1 voice model, even as xAI countered with Grok 4.5 specifically tuned for coding and autonomous agents. Meta entered the fray by launching both Muse Image and Muse video, claiming immediate ranking dominance, while ByteDance quietly strengthened its position with the release of Seedream 5 Pro. Meanwhile, Motion Previs Studio v4 integrated MCP support to streamline AI film pipelines, and Vidu S1 debuted real-time interactive digital humans that blur the line between simulation and presence.

This week feels less like a race to the top and more like a scramble for the most efficient path through the noise. Google opened its Gemini Omni Flash API while simultaneously releasing the budget-friendly Nano Banana 2 Lite, even as Adobe pushed Firefly Graph into enterprise node workflows. Anthropic quietly restored global access to Claude Fable 5 while Alibaba banned Claude Code internally, creating a stark contrast in corporate trust. Meanwhile, Slava Sexton shipped his local ComfyUI Agent Kit, allowing users to control pipelines via voice commands.

This week was a reckoning with access: the labs shipping the boldest tools also drew the tightest walls around them. OpenAI locked down GPT-5.6 Sol under government pressure, Adobe acquired Topaz Labs, and Figma turned its canvas into a full-stack playground with code layers and WebGPU shaders. Google revoked open Gemini API keys, Anthropic embedded Claude into Slack, and Sakana AI launched Fugu to orchestrate multiple models. ByteDance pushed hardest on video, shipping Seedance 2.5 with 30-second clips and native 4K.

This week was all tooling and money. Epic shipped Unreal Engine 5.8 with Mesh Terrain and MetaHuman crowds, Adobe put AI assistants inside Photoshop and Premiere, and the open-source side moved fast: Boogu's 10B image model, LTX 2.3's no-terminal LoRA trainer, and GLM-5.2 stretching context to a million tokens. ByteDance split Seedance 2.0 into tiers built around camera control, and Lionsgate backed the whole shift with a $5.3B stake in AI video.

This week AI moved in two directions at once: major launches on one side, abrupt shutdowns on the other. Anthropic shipped its most capable model yet in Claude Fable 5, only for fresh export controls to pull it offline outside the US days later. Meanwhile the creative stack kept building, with Apple threading a conversational Siri through its devices and Krea adding real-time sliders, while Luma's Ray 3.2 update landed softer than VFX artists had hoped.