Issue #26 · Friday, August 21, 2026 · 8 stories

🎞️ Edit Video by Talking to It 🏷️ Apple Music Punts AI Labeling to the Record Companies

Monday turns out to be peak creation day, and ComfyUI now hands your node graph to an agent.

Karate Girl did the rounds this week and it actually holds up, which comes down to two fairly boring paragraphs buried in the prompt. WasifAI told the camera to lag and overshoot like a real operator chasing the action, then locked the physics so every spark has to come from friction and contact rather than magic. Strip those two lines out and you get cartoon energy beams. Higgsfield is betting on the same idea by handing out the prompts behind a full feature film, and further down there is the duller news: 38.5% of July's music releases involved AI, and Apple Music has decided the labeling is the record companies' problem.

NEWS

38% of New Music Uses AI, Platforms Ban Undisclosed Tracks · 4 min read

A SubmitHub study of over a million July 2026 releases found 38.5% involved AI, with 23.2% fully AI-generated and 15.3% AI-assisted. Bandcamp and Beatport now restrict AI music, TIDAL withholds royalties on fully AI-generated tracks, and Spotify is rolling out AI labels. SubmitHub's detector is being adopted by Bandcamp, Traxsource, and German rights body GEMA, which recently won a landmark case against Suno. Deezer reports AI music exceeds 50% of daily uploads. Non-disclosure now carries real commercial risk as platforms enforce disclosure requirements.

Apple Music's AI Labels Depend on Labels Self-Reporting · 4 min read

Apple Music will display 'Made With AI' labels on tracks later this year, but enforcement relies on record labels and distributors tagging their own content via metadata. Apple does not detect AI content itself, unlike Spotify's independent review and Deezer's in-house detection. The EU AI Act requires generative AI services to embed machine-readable markers by December 2, which could give distributors a technical way to spot AI tracks, but enforcement will vary by platform.

Adobe Firefly Audio Ships Music, Speech, and SFX Generation · 4 min read

Adobe has made its Firefly audio generation generally available, adding three tools: Generate Music, Generate Speech, and sound effects. The music tool creates commercially safe tracks matched to video length and mood, with fine-tuning for brand fit. Speech converts scripts into voiceovers with control over voice, cadence, and emotion, and notably includes an ElevenLabs option. Sound effects are generated from prompts that follow the timing and energy of the content. All audio is production-quality and requires no separate subscription, making it a low-friction addition for video creators.

Free AI Filmmaking Academy Shares Prompts Behind 'Hell Grind' · 6 min read

Higgsfield AI launched a free academy that releases the prompts and workflows behind its AI feature films 'Hell Grind' and 'The Cully Hill Boys.' The curriculum covers characters, environments, camera movement, lighting, and continuity, with certification at the end. The company positions this as teaching AI integration into existing production pipelines, not replacing them, and reports 390 Fortune 500 companies and 170,000 corporate accounts on its Cinema Studio platform, which has a $500M annual run rate.

TECHNICAL

Free Prompt Lets You Edit Video by Talking to ChatGPT · 1 min read

A tutorial demonstrates a workflow for editing video through spoken commands: connect a clip to ChatGPT via the Borumi tool, then instruct cuts, layouts, zooms, and Remotion motion graphics. The creator trains a reusable editing skill and shares the prompt for free, while noting where the AI falls short. The value is the step-by-step integration and the reusable prompt, though it depends on a third-party tool that may need adaptation.

ANALYSIS

Monday Is Peak AI Video Creation Day, Virginia Leads at 5.5x US Average · 13 min read

Pictory's analysis of 1.5 million videos from its own platform finds Virginia users create 5.5x the US average per person, Oregon leads AI image generation at 35x the average, and Monday is the peak creation day in 12 of 15 large states. The data also shows Pennsylvania leads in voiceovers and avatars, and that consistent voice usage correlates with higher recognition. Single-platform data means the state rankings are directional, not causal, but the Monday peak and batch-production pattern are worth testing in your own workflow.

TOOLS

ComfyUI's official MCP hands the node graph to your agent, with beta caveats · 1 min read

ComfyUI shipped a first-party MCP server, so Claude or Codex can search nodes, models and templates, assemble a workflow, run it and repair it without you learning the node graph. There are two connections. The hosted one at cloud.comfy.org needs nothing installed and gives new users 5 free runs, after which generation requires an active Comfy Cloud subscription. The local server is open source, drives your own install for free, and needs Python 3.10 or newer plus comfy-cli, though partner models still execute on partner infrastructure and spend credits. Comfy is candid about where it breaks: this is public beta, workflow metadata does not always embed properly, complex multi-node workflows still need hand refinement, and outputs come back as shell downloads rather than inline. Treat it as scaffolding and repair for graphs you would otherwise wire by hand, not as unattended production.

Cool Video of the Week: the camera and physics prompt behind WasifAI's Karate Girl · 1 min read

🎬 Cool Video of the Week
This week's pick is Karate Girl from WasifAI, one continuous 30 second generation made with Seedance 2.5 at 1080p on OpenArt. The video gets better when you go into detail about the camera and the physics to make it look realistic, so here are those two sections of his prompt.

Clip and prompt by WasifAI on X. Tap the image to watch.

CameraClose, kinetic, reactive handheld pursuit; tracking, whip-pans, low-angle chase, OTS, facial inserts, rapid focus shifts. Camera moves with her but cannot perfectly anticipate acceleration; brief lag/overshoot enhances realism. Impact shake only from physical contact/discharge reaching lens. Action readability > chaos.
Physics lockStatic is friction-generated by extreme velocity/contact. Physical impact → charge buildup → branching discharge → dust/paper/fibers react. Never intentional energy projection, fireball, beam, aura, teleportation, or magic.