Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
MiniMax shipped Hailuo 2.3 with significantly improved physical motion rendering, complex character action fluidity, and sharper micro-expressions — plus better anime, illustration, and ink-wash stylization. A new "Fast" variant cuts batch generation costs roughly 50% versus Hailuo 02. Crucially, standard-tier pricing stays identical to the prior release, making this a free quality upgrade for existing users.
Google pushed a Flow update June 5th bringing native synchronized audio to all generation features — Ingredients to Video, Frames to Video, and Extend all produce matched sound now. Veo 3.1 Lite also landed in the Gemini API at under half the cost of Veo 3.1 Fast at the same generation speed. The platform has crossed 275 million videos generated since launching five months ago.
ComfyUI v0.24.0, rolling out through June 8th, introduces a unified Desktop app managing local, remote, portable, and cloud instances from one interface, plus day-zero Ideogram 4 support. The headline change is git-based auto-updates — the app now pulls ComfyUI updates whenever a release is tagged upstream, ending the separate desktop release cycle lag. Comfy Org raised $30M in April and the cadence of updates is reflecting it.
Every paid Runway plan now bundles Veo 3.1, Kling 3.0 Pro, and Seedance 2.0 as included third-party models — no separate accounts or API keys required. Runway also launched a conversational Agent in May for iterative creative direction. Note: the "Unlimited" plan transitions to "Max" for new subscribers now; existing users switch September 1, 2026.
Alibaba's Wan 2.7 — a 27B-parameter MoE open-weights model under Apache 2.0 — has reached full pipeline maturity in ComfyUI this week. It leads the Artificial Analysis text-to-video leaderboard at Elo 1762, well above prior leaders. The community technique gaining traction is its "9-grid" method: nine reference images in, one continuous character-consistent video out, no fine-tuning required.
TikTok now uses C2PA Content Credentials for automated AI content detection — if your generation tool embeds C2PA metadata (ComfyUI, Runway, and others do), TikTok can flag your video without a self-report. Three violations triggers approximately a 60% reach reduction for 30 days plus Creator Fund suspension. Meta still relies on self-declaration; that asymmetry creates real risk for creators pushing the same asset across platforms.
YouTube is actively demonetizing channels producing mass quantities of near-identical AI-generated content. Creator economy analysts put the threshold at roughly 60-70% AI production effort with meaningful human creative judgment carrying the rest. Current RPMs: AI tutorial and tech content is running $8-$15; AI short film and creative content is sitting closer to $2-$6.
OpenAI shut down the Sora app April 26th after approximately $1 million per day in compute costs against $2.1 million in lifetime revenue. The API stays live through September 24, 2026, then it's gone. Any production pipeline or automation built on the Sora API needs active migration now; Veo 3.1 and Wan 2.7 are the natural landing targets.
Banodoco open-sourced Hivemind in May — a live dataset of over one million Discord messages from AI artists and engineers discussing open video and image model techniques, structured as an agent-queryable skill. Coding agents can pull community best practices on generation workflows in real time. For prompt engineers doing deep technique research, it's a meaningful signal source that goes beyond static documentation.
