Black Forest Labs teaches video AI how to run robots

Good morning, {{ first_name | AI Enthusiasts }} and welcome to the 6,243 new readers who joined us yesterday. Black Forest Labs has made a name for itself in the world of AI imagery. Its latest model can create images, video and audio, and is currently running on Audi’s production lines.

FLUX 3, the company’s “visual intelligence” model, produces realistic video by simulating how the world actually moves. This is the same skill that currently helps us train and pilot machines in the physical world.

FLUX 3 extends from video to factory robots
Rowan’s Corner: How we build AI intuition at The Rundown
Edit videos faster with AI (for non-editors)
OpenAI launches dueling audio upgrade for Anthropic
4 new AI tools, community workflows, and more

black forest lab

Image source: Black Forest Labs

overview: Only at the Black Forest Research Institute opened Early access is now available to FLUX 3, a “visual intelligence” AI that generates video, images and synchronized audio, and the German institute is also incorporating it into FLUX-mimic, a robotic control variant for Audi’s production lines.

FLUX 3 produces 20 second clips in native audio with features such as multilingual dialogue, typography, and style/letter compliance.
The model can process text, image, and video inputs, and BFL’s tests show its output outperforms rivals like Runway, Kling, and Grok Imagine.
The complete deployment includes video, images, partner-specific action models, as well as open-weight FLUX 3 Dev sized to match factory hardware.
FLUX imitation, built Zurich-based imitation robotics is used to learn new factory tasks from about 30 minutes of demonstration data, rather than the usual 30 hours or more.

Why it’s important: BFL typically moves at a slower pace, but the integrated system appears to result in longer audio-synced videos and consistent visuals (as shown below) cheerful 1) It reminds me of the Sora era (in a good way). In the case of mimics, visual models continued to invade the physical world and became one of the key cogs in robotics.

With AWS

overview: AWS re:Invent will host more than 2,200 sessions in Las Vegas from November 30th to December 4th, 70% of which will be hands-on.

Tackle real problems in 300+ workshops and 250+ builder sessions
Debugging with your peers with over 125 code talks
Whiteboard your architecture with 350+ chalk talks with the engineers behind our services

rowan’s corner

Image source: The Rundown

Rowan: Last week I broken The most important skill in the AI era is AI intuition, knowing when to use AI and when not to use AI. This week, I’d like to dive deeper into how we’re actually applying it across our teams at The Rundown.

The graph above is a measure of AI fluency by role, mapping unacceptable, expected, and exceptional AI usage. Even for AI-first startups, frameworks like this set the standard with real-world examples.

Other rules from AI Fluency Notes:

AI is an amplifier, not a crutch. If AI makes your job worse, it’s your fault. The excuse that “AI did it” is never acceptable. Conversely, refusing to experiment means not being able to keep up with the best talent and AI agents.
Everyone spends two hours a week finding the boring part of their job, building a workflow, and sharing their accomplishments in the #ai-show-and-tell channel.
Everyone receives $200 per month for their AI tools. The only condition is that you must name the problem you are trying to automate. Not to control spending, but to ensure that all experiments start with real problems.

The goal behind everything is rep. AI intuition works similarly to business intuition. No one is born with AI intuition. Build by pushing the tools to their limits until it becomes instinctual when to use AI.

If your team hasn’t set such standards yet, we highly recommend it. here it is downloadable templates You can clone my setup.

AI training

overview: In this guide, you’ll learn how to use AI to edit videos, just like we do at The Rundown, and teach ChatGPT to do it hands-free.

Open Descript’s desktop app and record your camera and desktop if you want.
After recording is complete, duplicate the project before editing
Replication uses the AI tools “Remove Filler Words” and “Reduce Word Gap” to remove filler words and remove delays greater than 1 second. Turn on “Avoid Harsh Cuts” so your edits sound natural
Run Clarity Edits on Low and see the AI’s suggestions.

Pro tip: Connect ChatGPT or Codex’s Descript plugin and teach the editing process for hands-free editing.

Presentation by strand agent

overview: Your agent harness can have a bigger impact than the model you’re running. Strands Agent allows you to build your own agent harness and have end-to-end control without lock-in.

Run on any model, on any cloud, anywhere.

Monitor, modify, and debug using hooks
Deliver results coupled with guardrails that catch mistakes before they happen
Replace vulnerable scripts with tools that adapt to process changes.

open eye & antropic

Image source: @OpenAI and @Claudeai in X

overview: OpenAI and Anthropic shipped voice upgrade Around the same time, OAI migrated GPT-Live’s audio to its desktop app and Codex, and Anthropic connected Claude’s audio to its smart models and connected apps.

OpenAI’s ChatGPT Voice joins desktop apps on paid plans on macOS and Windows, allowing users to have voice conversations with agents in Codex and ChatGPT Work.
Custom hotkeys invoke voice mode on any app using Bloomberg. report That same technology could power OAI’s future home speaker hardware.
Claude’s voice mode now runs on Opus and Sonnet, as well as smaller Haiku models, and can also connect to apps like Gmail and Slack.
Claude has been delayed in speaking out of turn and has recently launch GPT-Live promotes the flow of listening, thinking, and responding at the same time.

Why it’s important: Both labs shipped something useful, but not what many users were looking for after the Claude 5 Opus drop and rumors of OAI speed upgrades by Cerebras. Despite the upgrades, Anthropic feels like it’s lagging behind its biggest rivals in the voice space, which continues to grow as the primary way it uses AI.

🎥 flux 3 – 20 seconds generated BFL multimodal AI
💬 poke – A proactive personal agent that resides within your text messages
🗣️ GPT Live – OpenAI’s more natural voice models now on desktop and Codex

cognition announced With the acquisition of The Interaction Company, the company’s text message-based AI agent Poke is being built alongside Cognition’s Devin agent.

US AI chip startup “Etched” safe Just weeks after coming out of stealth, the company has raised $300 million in new funding, and the company is now valued at $10.3 billion, bringing its total funding to over $1 billion.

Canadian mathematician Jacob Zimmerman teeth joining OpenAI tackles AI safety, the news was announced on the same day as his Award 2026 Fields Medal.

microsoft super intelligence team dropped Two in-house models with MAI-Image-2.5-Pro as a new image AI and MAI-Voice-2-Flash as a faster and cheaper voice engine.

runway launched Development platform media router. Automatically select image, video, or audio models for each request based on user preferences for cost, quality, and speed.

In each newsletter, we’ll show you how our readers are using AI to work smarter, save time, and make their lives easier.

Today’s workflow comes from a reader james Lancaster, Pennsylvania:

“I found a professional DJ controller on Facebook Marketplace that had been damaged by a water spill and wouldn’t turn on. I copied and pasted all of my conversations with the seller into Claude, which created a detailed risk assessment and price negotiation guidance. I ended up purchasing the unit for just $800.

I have uploaded the PDF service manual to Claude. Claude guided me through the professional disassembly and diagnostic process. I had uploaded a photo of the circuit board and Claude was explaining the exact terminals where voltage measurements should be made. The defective component turned out to be a $30 sensor strip inside the jog wheel. After replacing the parts and reassembling the unit, it powered up immediately.

Thanks to Claude, you can now get a $2,000 professional DJ controller for just $800, $30 in replacement parts, and a few hours of DIY tech service. ”

How do you leverage AI? please tell me here.