Amazon and LinkedIn tighten trust while Hailuo 03, PAL Maker, API cost math, and VOTW push creation.
An investigation reveals that supplement company Rosabella used AI-generated influencers on TikTok to sell products recalled by the FDA for salmonella contamination. The operation employed fake doctor avatars and stolen scripts, paying creators over $400,000 monthly in commissions. A competitor's lawsuit alleges false advertising and a misinformation campaign.
Amazon now requires third-party sellers to label product images containing photorealistic AI-generated people, adding a shopper-facing indicator. The policy follows New York's synthetic performer disclosure law, with penalties up to $5,000 per violation. Sellers must audit catalogs and embed IPTC metadata tags before upload.
MiniMax released Hailuo 03 (H3) after nine months, surprising with native audio, 15-second 2K generations, and an Omni mode that accepts up to 12 image references plus video and audio. Early tests show strong dialogue, lip sync, stable fight choreography, and a 7,000-character prompt limit. Pricing estimated at roughly 50% under Seedance 2.0. Currently in early access.
LinkedIn is testing a reporting tool that lets users flag posts as 'seems like AI slop,' which privately notifies the poster and reduces algorithmic reach. The company also removed its 'rewrite with AI' button in favor of a proofreading-only feature. A study found over 40% of longform posts on LinkedIn are fully AI-generated. For creators publishing AI content, this signals tighter enforcement and reduced organic visibility for AI-generated posts.
The AI video API field has three layers: model aggregators (per-second), workflow APIs (per-render, handles assembly), and publishing APIs (social posting). The article argues no single vendor covers all three, so developers should pair a workflow API like Syllaby with a publishing API. Cost comparison: $0.02-$0.08 per second for raw models vs per-render workflow pricing, with async lifecycle via webhooks, and workflow APIs save engineering time on assembly and maintenance.
Tavus launched PAL Maker, a no-code platform for building AI video agents that can hold face-to-face conversations, remember users, and integrate into websites or Google Meet. Users describe the desired PAL in plain language, and the system builds it with customizable faces, voices, and knowledge base grounding. Targets customer support, onboarding, and team meetings. Marketing-heavy but appears functional.
🎬 Cool Video of the Week
This week's pick comes from Dreaming Tulpa, who tested how far FLUX 3 can go with a deliberately loose dialogue prompt: ranting about ai. The clip works because it does not feel over-directed. The body language, handheld framing, outdoor light, and slightly messy delivery all land in the uncanny sweet spot where the realism is convincing before you start hunting for seams. That is the practitioner takeaway: vague prompts are starting to produce usable performance, not just pretty motion.
Clip by Dreaming Tulpa. Tap the image to watch on X.