Two tools earn your attention this week: Flux 3 is the first Black Forest Labs model to generate video with native audio, up to 20 seconds, and LTX-2.3's Clean Plate LoRA rebuilds the background after it removes people and vehicles, no mask required. Around them the business news got heavier, from Netflix's 300 AI titles and Suno's breach and lawsuits to a German AI copyright ruling and Nvidia's millisecond deepfake detector. Stick around for the Lyrebirds reel at the end, it earned the closer.
NEWS
Black Forest Labs announced Flux 3, its first model that generates video with native audio, with clips up to 20 seconds. It is one multimodal backbone across image, video, audio, and action prediction, covering text-to-video, image-to-video, video-to-video, multilingual dialogue, and chaining of clips into multi-shot sequences. Video is in early access now, image follows in the coming weeks, and an open-weight Flux 3 Dev is planned. BFL's own preliminary evals put it ahead of Runway Gen-4.5 in 77 percent of comparisons and Luma Ray 3.2 in 93 percent, though those are self-reported and the release is still limited, so treat the numbers as a starting point rather than a verdict.
Netflix disclosed in its Q2 2026 earnings that generative AI was used in roughly 300 of its programs this year, primarily in production and postproduction. Co-CEO Ted Sarandos framed AI as a cost saving and creative tool that makes possible shots otherwise impossible within budget and time constraints, but emphasized it is not a replacement for human artists. The disclosure signals that major streaming platforms are integrating AI into mainstream production workflows, validating AI tools for content creation.
AI music generator Suno suffered a data breach affecting over 55 million users. Stolen data includes names, addresses, emails, and partial payment card details. The breach also exposed Suno's source code, revealing how the company allegedly scraped songs and lyrics from streaming platforms to train its models a practice at the center of ongoing copyright lawsuits. For builders, the incident highlights both the security risks of handling user data and the legal exposure from training data sourcing.
A German higher court ruled that an AI-generated comic version of a dog photographer's underwater photo did not infringe copyright. The AI altered framing, perspective, and lighting, retaining only the unprotected subject matter. The decision clarifies that transformations changing the original's artistic choices may avoid infringement, setting a legal precedent in Germany for AI media creators.
Nvidia launched a microservice that identifies AI generated video frame by frame. On uncompressed 1080p footage processed on an RTX GPU, it reaches 92% accuracy in about 22 milliseconds. Compression reduces that accuracy. The tool assigns a probability score per frame. For builders shipping AI video content, detection is now fast enough to run at scale.
TOOLS
Lightricks released an open-source IC-LoRA for its 22B LTX-2.3 video model that takes a source clip and regenerates the same shot with people, pedestrians, and vehicles removed, rebuilding the empty background behind them while keeping architecture, ground markings, and foliage intact. There is no mask step: it runs video-to-video in ComfyUI, was trained at 1024x576 and 576x1024 at 49 frames, and was validated up to HD. The practitioner payoff is skipping a reshoot when an unwanted subject wanders into frame, though full-frame regeneration means you are trading the original pixels for the model's reconstruction.
QUICK HITS
🎬 Cool Video of the Week
This week's pick is from Lyrebirds, an AI film studio that treats these tools as instruments rather than authors. A man rides a hand-painted scooter down a dusty village road with two chrome robots riding pillion and a dog balanced up front. The whole shot is AI-generated, and the tell is how well it holds together: the character stays consistent, the dust moves right, and the light sits still across the frame. Tap the image to watch it on Instagram.
Clip by Lyrebirds. Tap the image to watch on Instagram.