Veo 3 gets camera control

In partnership with

Welcome to VP Land! Control is the theme this week—from Veo 3 camera adjustments to real-time motion guidance, AI video tools are getting more interactive.

In today's edition:

Veo 3.1 adds camera controls
Comfy Cloud opens to everyone
Nano Banana 2 leak briefly surfaces (maybe?)
MotionStream enables real-time AI video

Google Veo 3.1 Adds Camera Controls

Google rolled out camera position and camera motion controls in Veo 3.1. The update lets users adjust camera direction and movement after generation.

— # (#)

Post-generation camera controls let you modify camera direction within already-generated Veo videos
Camera motion controls enable changes to pans, tilts, and simulated camera moves after the initial video is created

The key limitation: These controls only work on videos already generated within Veo 3.1. You can't upload external footage and apply Veo's camera system to it.

SPONSOR MESSAGE

Read newsletters, not spam

Tired of newsletters vanishing into Gmail’s promotion tab — or worse, being buried under ad spam?

Proton Mail keeps your subscriptions organized without tracking or filtering tricks. No hidden tabs. No data profiling. Just the content you signed up for, delivered where you can actually read it.

Built for privacy and clarity, Proton Mail is a better inbox for newsletter lovers and information seekers alike.

Newsletters without the noise

Comfy Cloud Opens Public Beta

Comfy Cloud, the browser-based version of ComfyUI, has opened its public beta after operating under a waitlist. The service lets users run ComfyUI workflows entirely in the cloud, removing the need for local hardware or installation.

According to the announcement, here's what you get:

NVIDIA A100 GPUs powering your workflows, with access to over 400 open-source models and 17 preinstalled community extensions
$20/month pricing that includes $10 in credits for "Partner Nodes"—third-party models like Sora, Veo, and Google's nano banana (Gemini 2.5 Flash)
8 GPU hours per day limit during beta to manage capacity (not billed), with plans to increase limits for heavier users
Runtime-only billing after beta ends—you won't be charged while idle or editing workflows

The platform is actively scaling GPU capacity during beta. The roadmap includes features like custom model uploads, LoRAs support, API deployment for workflows, parallel workflow execution, and team collaboration tools.

What matters: If you've been curious about ComfyUI but didn't want to deal with local setup or hardware requirements, this lowers the barrier significantly.

An unverified leak suggests Google's Nano Banana 2 image generation model surfaced briefly on select platforms before being pulled. According to filmmaker Roberto Nickson on X, a small handful of users accessed what appeared to be an uncensored, older checkpoint of the model, with results he described as "pretty insane."

What's claimed (and what's missing):

Super sharp text generation: some of the posted images supposedly from the model showed extremely accurate chyrons and text in fake screen grabs from CNN, and a full YouTube interface was generated on top of a Windows background with no warping.
Access was extremely limited: Only a few users reportedly accessed the model during a brief window on unspecified platforms
Allegedly uncensored: The leaked checkpoint apparently lacked the content filtering typical of Google's public AI releases, with photorealistic images of public figures posted online
Not the final version: The model was reportedly an older checkpoint, and the eventual public release will likely differ significantly

No word on an official timeline or release.

Source: Roberto Nickson on X

MotionStream: New AI Video Tool Lets You Guide Motion

A new research tool called MotionStream addresses one of the most persistent frustrations with AI video generation: real-time control.

The system processes motion guidance in real-time, giving you quick feedback as you define paths directly in the video canvas.

What makes this different:

Draw motion paths directly - Frame-accurate control over subject movement, combined with text prompts for scene content.
Real-time feedback - The diffusion model architecture delivers previews in seconds as you adjust controls, dramatically reducing iteration time compared to batch-based workflows.

Tom and Dominic of The Rest Is History tour Disney Studios and Disneyland with CEO Bob Iger.

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

⚡ Wan 2.2 Animate just rolled out a massive upgrade, delivering 4× faster inference and sharper visuals for just $0.08 per second at 720p.

🎨 Azornes’ ComfyUI-Resolution-Master delivers granular control over AI image resolution and aspect ratio with an interactive canvas and model-specific presets.

🪐 New from Bytedance: InfinityStar, a unified autoregressive model for visual generation that synthesizes images and video from spacetime data.

🎬 Bytedance launches Video Upscaler on fal, enabling professional-grade video upscaling to 1080p, 2K, and 4K.