← Back to archive
Issue #47 · Friday, October 16, 2026 · 7 stories

🎤 Taylor Swift Trademarks Her Voice In Australia, 🏷️ YouTube Starts Labelling AI Music

What a talking avatar costs per minute, and two image models that got cheaper and faster this week.

The legal risk in AI content is moving from your pictures to your voice. Taylor Swift's spoken-phrase trademark was accepted in Australia on 12 October, and trademark law reaches sound-alikes, not only copies. Platforms are taking the softer route: YouTube Music now labels AI songs rather than blocking them. A label will not protect an ad that borrows a famous voice. Also inside: avatar costs per minute, FLUX 3 Klein and Krea 3.
NEWS

YouTube Music now tags AI-made songs with an AI label under the track name · 2 min read

Since 14 October, tracks where AI played a significant role, fully generated songs or ones with AI vocals, carry an oblong AI label under the title. Tapping it explains how AI was used for listeners. The same label is coming to the main YouTube player. YouTube has not said how it decides, so expect your AI vocals to be labelled whether or not you disclosed them.

A Suno-made song with human lyrics reached #3 on South Korea's Melon chart · 1 min read

'The Adult I Dreamed Of', credited to Meomuru, pairs human-written lyrics with music generated in Suno. Since its June release it has climbed to #3 on Melon and #2 on YouTube's weekly South Korea chart with 3.2 million weekly views, driven by use in social posts. It now carries YouTube's new AI tag, so a label did not stop a chart run.

Taylor Swift files voice trademarks in Australia, the UK and EU, and Australia already accepted one · 4 min read

Swift's company has filed sound marks for the spoken phrases "Hey, it's Taylor" and "Hey, it's Taylor Swift", following US filings in April. Australia accepted its application on 12 October, opening a roughly two-month opposition window. Trademark law covers imitations that are confusingly similar, not only exact copies, so a sound-alike celebrity voice in an AI ad is now a trademark risk as well as a likeness one.

TECHNICAL

Talking avatars cost $2.32 a minute on HeyGen and $8.26 on Higgsfield · 17 min read

Synthesys, which sells a competing avatar tool, worked out per-minute costs from each plan's credits: HeyGen Avatar V on Creator burns 48 credits a minute, about $2.32, and Higgsfield Speak 2.0 on Plus about 168, roughly $8.26. Only Higgsfield was actually tested so far. Both platforms' terms license your uploads for model training, and HeyGen's free plan bars ads and client work.

TOOLS

FLUX 3 Image Klein runs on an 8 GB GPU and is free for commercial use under $5M revenue · 1 min read

Black Forest Labs released FLUX 3 Image Klein on 15 October, a 5.6B-parameter open-weights model for generating and editing images. The INT8 build runs on 8 GB GeForce RTX cards and makes a 1MP image in under 9 seconds on a 4070. The catch: free commercial use stops at $5M annual revenue, above which you pay for a licence or the API, from $0.009 per 1K image.

Krea 3 returns most 2K image edits in under 8 seconds at $0.06 each · 1 min read

Krea's new editing model returns most edits in under 8 seconds at native 2K, against 33 seconds for Nano Banana Pro and 90 for GPT Image 2 in Krea's own comparison. It costs $0.06 per 2K image or $0.03 at 1K, takes up to six reference images, and is live on krea.ai plus fal, Replicate and OpenRouter. Edits are text-prompt only for now; on-image markup comes later.

QUICK HITS

Cool Video of the Week: a hand turns to molten copper on a city street · 1 min read

Posted by @Enaya_ai on 14 October and made with Seedance 2.5: a hooded figure sprawled on a city street reaches out a hand that has turned to molten copper, pooling on the asphalt. It runs 30 seconds and has passed 3.3 million views. The video plays for logged-out readers; the replies need an X login.