← Back to archive
Issue #53 · Friday, October 30, 2026 · 7 stories

🎤 Elvis's Song Owners Sue Anthropic For US$48.6M 🎵 Suno Now Makes Songs Inside Claude

An Oscar-qualifying short trained AI on its own concept art. A Blender add-on rigs from one prompt.

Polish is turning into a tell. The BBC says not a single frame of Blue Planet III is AI, and it proves it by keeping the plankton, murky water and stray boats it used to cut. That is bad news for anyone whose pitch is flawless footage, and it is why Dear Upstairs Neighbors matters: it reached Oscar eligibility by training models on its own concept art, so the look belongs to its artists, not to a default model. Platforms will not sort this out for you either, since VRChat just declined to add AI labels after 21,595 players could not agree on one. This week's tools, a free Blender rig-and-texture pipeline and Suno inside your chat, are only as good as the taste you bring.
NEWS

Elvis Presley's song owners sue Anthropic for US$48.6M, saying Claude reproduces their lyrics on request · 8 min read

The owners of 274 songs Elvis Presley recorded, led by Authentic Brands Group's ABG EPE IP, sued Anthropic on 23 October in the Northern District of California. The complaint says Claude was trained on scraped lyrics with songwriter credits stripped out, and that it outputs copies, and new works based on them, when users ask. The pleaded figures come to just over US$48.6 million, and the plaintiffs want unauthorised copies destroyed, including those in model weights. Universal Music Publishing is already suing Anthropic in two separate cases. If you draft song lyrics in Claude, check them against the originals before anything goes out.

95% of 21,595 VRChat players want AI labels. VRChat is adding none, because they could not agree on one · 9 min read

VRChat surveyed 21,595 players in a week, and 95% want some kind of label for AI use. Their answers produced 3,141 different labelling schemes, though, and no single one got more than 10%; only 33% would deliberately avoid an "AI-assisted" world. On 28 October VRChat said it will not add an "AI-generated" label or AI filters, because some creators would label honestly, others would not, and it expects sleuthing and harassment to follow. Misleading thumbnails are now against the rules instead, likes and dislikes on worlds are due by the end of 2026, and a "Human-authored" mark is under consideration. If you build worlds with AI, fix the thumbnail first.

TECHNICAL

An animated short finished with AI models fine-tuned on its own concept art has qualified for the 2027 Oscars · 4 min read

Connie Qin He's Dear Upstairs Neighbors qualified for Oscar consideration through a one-week theatrical run. The team made the storyboards, designs and 2D and 3D animation by hand, then used generative models fine-tuned on their own concept art to produce the final painterly imagery. He says the models picked up more than brushwork and palette: they learned the film's two-point perspective and its habit of drawing sound as brushstrokes. The order is the part worth copying. The art came first, and the model was trained to finish it, not to invent it.

ANALYSIS

Blue Planet III keeps murky, plankton-filled shots on purpose, as proof that no frame is AI · 2 min read

"No gen AI was used in this series. Not a single frame," the BBC's Jack Bootle said at a press screening for Blue Planet III. Executive producer Elizabeth White says shots that were once cut for cloudy water, plankton, or a boat or buoy in frame now stay in, because AI "makes us want to be even more authentic." The series was filmed over four years across five oceans. For anyone making AI footage, that is an uncomfortable signal: clean, flawless water is becoming the look audiences read as fake, and real productions now use imperfection as their proof.

TOOLS

StableGen 0.4.0 turns one prompt into a textured, rigged character in Blender, but its Qwen path is non-commercial · 9 min read

Version 0.4.0 of this free Blender add-on, out 28 October, adds automatic rigging: a text prompt becomes a mesh built by TRELLIS.2, rigged by SkinTokens and textured from views projected around it, ready to export as FBX for Unreal and Unity or glTF for the web. Mixamo animations retarget onto the rig, which makes it a quick previs source. It runs its models through ComfyUI and wants about 8 GB of VRAM for SDXL texturing, 16 GB or more for Qwen Image. Read the licence section before client work: Qwen Image 2.1, Bria FIBO 2.0 and the TRELLIS.2 Native texture mode are non-commercial; SDXL and Qwen Image Edit are not.

Suno now makes songs from inside Claude, ChatGPT, Gemini and Cursor, on your existing credits · 4 min read

Suno's new MCP server connects your Suno account to ChatGPT, Claude, Claude Code, Cursor and Gemini. You sign in once, with no API key, and ask the assistant for a song, a cover, an extension or a full track built from a voice memo; songs use the credits on your current plan and save to your Suno library. The useful part for video work is context, since the assistant can write lyrics and a style from a script or video plan already in the chat. Suno does not say how many credits a song costs this way, and some chat apps only allow custom connectors on paid plans.

QUICK HITS

Cool Video of the Week: an app founder pitches Seedance 2.5 cooking reels over US$4,000 chef videos · 1 min read

Split screen: a tray of glossy baked Nutella brioche beside a Nutella jar on the left, and a grid of Instagram cooking reels showing 7.8M to 20.5M views on the rightApp founder Ernesto Lopez posted this 29-second split screen on 29 October: a Nutella brioche reel next to a cooking account whose visible reels sit between 7.8M and 20.5M views. His pitch is that Claude Motion plus Seedance 2.5 can now make cooking content like this to market an app, where a chef or influencer used to cost US$4,000 a video. The revenue figures in his post are his own and come with no receipts. The format he is pointing at is the thing to study: marble backdrop, serif title card, overhead hero shot, a hand in frame. Posted by @ErnestoSOFTWARE.