AI Assistants Are Moving Into the Tools That Make Films, Games and News

Key points

• NVIDIA announced five new AI tools at SIGGRAPH 2025 in Los Angeles, running through 23 July.
• Adobe, Blender, Unreal Engine, Houdini and Affinity by Canva are adding MCP connections, letting AI agents work inside the software artists already use.
• NVIDIA's new Synthetic Video Detector analyses video frame by frame and gives newsrooms a confidence score indicating whether footage was created by AI.
• The Cosmos 3 Edge world model, designed to run on a local machine rather than a remote server, lets physical AI systems such as robots understand their surroundings.

Imagine telling your design software: "Find every layer with a missing texture and flag it for me." Until recently that sentence described a task a human had to do by hand. At SIGGRAPH, the annual graphics research conference running through 23 July in Los Angeles, NVIDIA showed how that kind of request is becoming routine.

The key technology is the Model Context Protocol, or MCP. Think of MCP as a standardised plug socket: creative applications install the socket and AI agents plug in. As we reported on 20 July, updates to MCP are already making it simpler for AI tools to connect to software without custom engineering. Once connected, an agent can read a scene, check files, execute repetitive tasks and hand results back to the artist, without the artist ever leaving their software.

Adobe is expanding MCP support across Firefly and Express. Blender, the free 3D modelling tool used by independent artists worldwide, has a lightweight MCP server that lets you talk to it in plain language. Unreal Engine now lets AI clients connect directly to its editor. SideFX is bringing MCP to Houdini 22, starting with character rigging. Affinity by Canva has connected the AI assistant Claude so designers can say things like "resize all these assets for Instagram" and watch it happen.

The through-line is human control. Agents handle dull production work; artists keep final say over every creative decision.

Should ordinary people care about any of this?

Yes, because it changes what you watch and play. Faster, cheaper production pipelines mean studios can produce more content. But it also means synthetic, AI-generated material is easier to make and harder to spot.

That's exactly why NVIDIA's second big announcement matters. The company introduced the Synthetic Video Detector, built as a NIM microservice (a small, self-contained piece of software a newsroom can plug into its existing systems). It analyses footage frame by frame and outputs a confidence score showing how likely a clip is to have been generated by AI rather than shot with a camera.

Editorial teams can use that score to flag suspicious footage, prioritise clips for human review or pull material from air while fact-checkers dig deeper. NVIDIA says the detector keeps working even after video has been compressed or re-encoded, the routine steps that strip out many other detection signals.

The tool doesn't replace journalists. It gives them one more data point when decisions have to be made fast.

NVIDIA also unveiled Cosmos 3 Edge, an open world model (software that helps a machine understand how the physical world works) designed to run locally on a single device rather than needing a remote data centre. That matters most for robotics and autonomous systems, where a connection delay can be the difference between safe and unsafe. Jensen Huang's recent Tokyo trip, which we covered on 19 July, centred on exactly this kind of physical AI push into robotics and manufacturing.

All five announcements were presented at the NVIDIA SIGGRAPH keynote on 20 July.

Common questions

Can the Synthetic Video Detector be fooled?

NVIDIA says it holds up after compression and re-encoding, but no detector is perfect. Newsrooms should treat the score as one signal among several, not a final verdict.