Issue #16 · Wednesday, July 29, 2026 · 8 stories

🪪 China Rents Faces by the Episode, and Fish Audio Goes Free Until August 31

Disney turns brand logos into TV spots, plus three changes to make before your next upload.

Chinese platforms now rent a real person's face for 99 to 500 yuan an episode, and GEMA is selling a fully licensed 178,000 file music dataset days before its Suno verdict lands. Disney is automating CTV ads from brand assets, Fish Audio shipped 83 language voice cloning with a free tier, and the analysis explains why disclosure, not quality, is what gets channels killed. In tools: a free Descript alternative, an LTX camera LoRA that names its own limits, and three continuity fixes.

NEWS

Face Licensing Platforms Feed China's 95% AI Microdrama Market · 2 min read

Chinese platforms ActID and New Claw let people license their faces for AI generated content, charging creators 99 to 500 yuan per episode or image. The platforms take a 10% commission, with contributors earning $15 to $700 per face. This model serves China's microdrama industry, where 95% of 128,000 releases in Q1 2026 used AI. Legal protections exist under Chinese biometric data law, but enforcement is weak. For AI video creators, this offers a scalable way to source consistent characters without traditional shoots, though misuse risks remain.

Licensed Music Dataset for AI Training Goes Live Ahead of Copyright Verdict · 4 min read

German collecting society GEMA launched PLAI, a fully licensed dataset of 178,000 sound files across 60+ genres, with Klangio as its first customer. The dataset bundles composition and master rights from a single source, but is only cleared for assistive AI tools that don't compete with the training music. GEMA awaits a copyright ruling against Suno on July 31 that could set licensing precedent. For builders working on AI music products, this offers a compliant training data option, with the caveat that generative models that compete with the source material are excluded.

Disney's AI Ad Studio Turns Logos Into CTV Spots · 3 min read

Disney opened a closed beta of Ad Creative Studio, an AI tool inside Disney Campaign Manager that takes brand assets like logos and product images and generates CTV video ads. The system uses multiple AI models to create variations targeted by audience, geography, and campaign goals, with human review and IP protections. For builders in ad tech or AI video, this is Disney's first move into automated ad production, potentially lowering the barrier for small and mid-sized businesses to run TV ads.

ANALYSIS

Trust Infrastructure, Not Quality, Is AI Video's Real Problem · 6 min read

An analysis argues the failure behind Sora's shutdown and YouTube's purge of AI slop channels is a lack of trust infrastructure, not video quality. The piece offers three concrete updates for creators: audit your disclosure practices against platform policies (YouTube now auto-labels AI content with high confidence), deliberately produce less than the ceiling of cheap tools like Veo 3.1 Lite allow, and name the human decision maker on every AI assisted piece before shipping. The argument grounds itself in real events including Sora's deepfake crisis and Google's own case studies of deliberate AI use.

TOOLS

Fish Audio Puts S2.1 Pro Behind a Free API Until the End of August · 1 min read

Fish Audio's anniversary post is the announcement behind this week's launch: a $52M seed led by Coreline Ventures and Capital Today, $21M ARR, more than 8 million users and over 2 million community voice models, built by a team that went from 3 people to 22 in a year. The part that changes your week is the offer attached to it. S2.1 Pro, the largest model, is free to developers via API through the end of August, with 50% off creator plans and three months free for migrations. The model claims 83 or more languages with native cadence, word-level emotion control across 15,000 natural language tags, more than 8,000 tokens per second on a single H200, and a 66% preference rate against competitors in blind listening tests. Note what the post does not do: it gives no per-character pricing to compare against ElevenLabs or Cartesia, only that inference optimisation dropped cost per request far enough to give the model away. Free until August 31 is a window to test in, not a cost you can plan a business on.

Free Browser Video Editor Edits by Changing Transcript Text · 1 min read

Rescript is a browser based video editor that lets you cut and rearrange footage by editing the transcript text, similar to Descript. It runs completely offline and locally, making it a privacy friendly option for creators. The tool is launching today on Product Hunt and available on GitHub.

New LTX LoRA re-angles footage you already shot, and warns the rotation falls short · 1 min read

Cseti's CrossView-Warp IC-LoRA for LTX-Video 2.3 takes an existing clip plus a numeric camera offset (azimuth, elevation, distance) and regenerates the scene from that viewpoint. A 3D orbit picker keyframes a full camera move across the clip, and the model is fed two references: a depth-warped copy that carries geometry, and the original that holds identity. The model card is honest about the catch. It steers the viewpoint rather than reprojecting geometry, so it routinely rotates less than you asked for. This is a v0.9 proof of concept trained on 548 real and synthetic camera pairs, and it needs the 22B LTX-Video 2.3 base under Lightricks' licence, so price the VRAM before you plan a shoot around a reshoot you cannot do.

Three Techniques to Fix Disjointed AI Video Clips · 2 min read

A tutorial breaks down three methods for maintaining continuity across AI generated video scenes: continuing generation from a previous clip, frame chaining from the final image of a scene, and locking character details, outfits, and locations. The creator also demonstrates using Codex with the Higgsfield MCP to gather references and organize projects. The advice is actionable for daily production, though the video includes affiliate links to Higgsfield.