AI film premieres at Sundance

In partnership with

Welcome to VP Land!

In this issue:

Google's Sundance AI film - Pixar vet directs with Veo and Imagen
~~Clawdbot~~ Moltbot - The open-source AI agent breaking the internet
KREA Realtime Edit - Live image editing as you type
LoRA vs Reference Images - Which workflow actually works?

Google's Sundance Double Play

Google premiered Dear Upstairs Neighbors at Sundance, a 6-minute animated short directed by Pixar veteran Connie He. The film follows a sleep-deprived woman whose noisy neighbors trigger increasingly unhinged hallucinations, rendered through fine-tuned versions of Veo and Imagen.

The workflow is what makes this interesting. Rather than generating shots from scratch, the team animated in Maya and TVPaint first, creating rough 2D/3D animation that AI then stylized frame-by-frame. It's a collaborative model where AI handles stylization while artists maintain control over motion and timing.

Video-to-video processing - Animators created the performance and composition; AI applied painterly styles to each frame. The underlying motion stays human-controlled.
Fine-tuned models - The team used customized versions of Veo and Imagen trained on specific visual styles, not off-the-shelf generation.
Pixar pedigree - Director Connie He brings traditional animation expertise, treating AI as a rendering tool rather than a replacement for artistic direction.

The template - This positions Google's approach as "artists + AI" rather than "AI replaces artists," a framing they'll likely push as the industry debates automation.

Google also announced $2 million in funding for an AI Literacy Initiative to train over 100,000 artists through Sundance Collab, partnering with The Gotham and Film Independent.

Read More to learn about the film's production workflow.

SPONSOR MESSAGE

Better prompts. Better AI output.

AI gets smarter when your input is complete. Wispr Flow helps you think out loud and capture full context by voice, then turns that speech into a clean, structured prompt you can paste into ChatGPT, Claude, or any assistant. No more chopping up thoughts into typed paragraphs. Preserve constraints, examples, edge cases, and tone by speaking them once. The result is faster iteration, more precise outputs, and less time re-prompting. Try Wispr Flow for AI or see a 30-second demo.

Start flowing free

Claude Plugs Into Your Tools

Claude can now interact directly with Slack, Figma, and Asana from within the chat interface, turning conversations into actions without switching apps. The shift from "AI that answers questions" to "AI that does things in your tools" keeps accelerating.

Anthropic uses Model Context Protocol (MCP) Apps to embed live, interactive instances of third-party tools inside Claude. Ask Claude to create an Asana project from your conversation, and it opens an actual Asana board in the chat where you can review and tweak before committing.

Figma integration - Generate flow charts, Gantt charts, or diagrams from text descriptions via FigJam. Claude interprets your prompt and creates editable visuals you can refine without leaving the conversation.
Slack integration - Search past conversations for context, then draft messages that Claude formats to match Slack's interface. Review before posting; no accidental sends to #general.
Asana integration - Convert brainstorming sessions into projects, tasks, and timelines. Claude structures the work, you approve and assign.

This builds on Cowork's computer control capability we covered previously. Anthropic is steadily expanding from answering to doing, from isolated chat to embedded workflow. Available now for Pro, Max, Team, and Enterprise users at claude.ai/directory.

If you've been on X this week, you've seen Clawdbot everywhere. The open-source AI assistant has exploded from zero to 60,000+ GitHub stars, with Andrej Karpathy and David Sacks both posting about it. One user put it bluntly: "Clawdbot is Jarvis. It already exists."

So what is it? Created by Peter Steinberger (founder of PSPDFKit), Clawdbot is a self-hosted AI agent that runs on your Mac, PC, or server and actually controls your computer. Unlike ChatGPT or Claude's web interfaces, it executes terminal commands, manages files, browses the web, and interacts with apps autonomously while you sleep.

Always-on messaging - Connect via WhatsApp, Telegram, Slack, Discord, iMessage, or Signal. Message your AI assistant from anywhere, get responses in the same chat.
Persistent memory - Remembers your preferences and past conversations across sessions. No context window resets.
50+ integrations - Gmail, GitHub, Spotify, Obsidian, Notion, Apple Notes. The AI can write and modify its own skills.
Proactive actions - Morning briefings, reminders, alerts when something you care about happens. It reaches out to you.

The hype is real, but so are the caveats. The GitHub repo has 500+ open security issues, and running an AI with full system access requires serious sandboxing. Cost-wise: free and open-source, but Claude API usage runs $20-50/month for moderate use, to $100s a day for heavy usage.

Plot twist: Anthropic just issued a trademark request forcing a rebrand (the name "Clawd" was too similar to "Claude"). It is now Moltbot.

KREA's Real-Time AI Adds Editing

KREA added Realtime Edit, which lets you upload an image and modify it with natural language while watching changes happen live. Type "make the sky more dramatic" and the image updates as you type.

Live editing - Lighting, color, composition, and element swaps update continuously as you refine your text input.
Any source - Works on photos, reference art, AI-generated images, or stills from footage.
Storyboard iteration - Pull a frame, describe changes, adjust until it matches what you want.

Currently in beta with gradual rollout.

Are LoRAs still necessary for capturing specific cinematography styles, or have reference images caught up? We put two workflows head-to-head trying to replicate the VistaVision look from One Battle After Another.

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

⚡ Black Forest Labs released FLUX.2 [klein] with sub-second generation on consumer GPUs, hitting 0.3 seconds on RTX 5090 for real-time image iteration without quality loss.

🚀 Luma AI's Ray3.14 update brings native 1080p, 4x faster generation, and 3x cheaper pricing, eliminating the quality-speed-cost tradeoff that's held back production use.

📉 Vimeo gutted its workforce after Bending Spoons' $1.38B acquisition, following the company's pattern of eliminating 75% of WeTransfer and all of Filmic's staff post-purchase.

🎬 Topaz Labs shipped 11 new enhancement models in their Fidelity update, including Wonder 2 for detail restoration and Proteus upgrades that eliminate the over-sharpened AI look in upscaled video.

🎤 Alibaba open-sourced Qwen3-TTS with voice cloning from 3 seconds of audio and 97ms streaming latency, offering an ElevenLabs-competitive option under Apache 2.0.

Addy and Joey explore Claude Cowork's desktop control, PixVerse real-time video, FLUX.2 [klein]'s sub-second generation, Apple's $13/month Creator Studio bundle, and ComfyUI's latest nodes.

Read the show notes or watch the full episode.

Watch/Listen & Subscribe

Spotify | Apple Podcast | YouTube