Runway launched Runway Agent on May 13, an agentic creative partner that takes a text description and reference images through concept, storyboarding, multi-scene generation with voiceover, dialogue, and music — outputting a finished, publishable video in minutes. It supports aspect ratio and duration config, iterative direction via conversation, and a timeline editor for final tweaks. Available now to all Runway users.
Video Rebirth (founded by ex-Tencent distinguished scientist Dr. Wei Liu) launched BACH on May 7, an AI video engine that turns a single idea into a 30-second multi-shot film with consistent character identity, directable emotions per shot, and professional camera language. BACH ranked #6 on the Artificial Analysis Video Arena before public launch, on par with Kling 3.0 Omni. Built on a proprietary Dual Diffusion Transformer architecture. Available at bach.art.
Within four days of Google I/O 2026, Adobe (May 20), Canva (May 19), and CapCut (May 21) all confirmed native integrations inside the Gemini app. Canva is already live — type @Canva in a Gemini chat to generate designs, search existing content, and convert Gemini-created images into layered editable projects. CapCut and Adobe integrations are announced but not yet live. Note: Google also shifted to compute-based usage limits on May 17, sharply reducing the AI Pro multiplier from 33x to 4x.
Runway shipped Gen-4 on May 3 with native audio generation (soundscapes, dialogue, and environmental effects synthesized frame-by-frame), new API hooks for hybrid pipelines with Veo3/Seedance/other models, and a reported 40-50% reduction in object pop-in versus Gen-3. This is a significant jump for Runway's competitive positioning against Kling 3.0 Omni and Seedance 2.0.
LTX-2.3 (March 2026) is now fully integrated into ComfyUI with 14 free workflow templates, native text-to-video, image-to-video, and audio-to-video in one pipeline. The 22B-parameter model supports 4K at 50fps with stereo 24kHz audio, 9:16 portrait mode, and improved text rendering. Weights are free on HuggingFace. ComfyUI v0.16 shipped Day-0 support — no custom node packs needed. A 10-second 1080p clip runs ~7-8 minutes on an RTX 4090.
Alibaba's ATH unit was unmasked as the creator of HappyHorse-1.0, a 15B-parameter model that anonymously climbed to #1 on the Artificial Analysis Video Arena for both text-to-video and image-to-video. It generates video and audio jointly in a single forward pass using a unified 40-layer self-attention Transformer — ~38 seconds for 1080p on a single H100. Published under Apache 2.0; full weights and GitHub repo are marked "coming soon" as of late May.
Instagram began testing a new "AI Creator" label on May 4 that appears on profiles, bios, and alongside posts and Reels in Explore. The label is entirely optional and Instagram has confirmed it does not affect ranking, distribution, or recommendations during the test. Meanwhile, the EU's mandatory AI content disclosure requirement kicks in August 2026 — creators with European audiences should prepare now regardless of Instagram's voluntary approach.
OpenAI shut down sora.com and all Sora mobile apps on April 26. The API (including Sora 2 and Sora 2 Pro endpoints) will go dark on September 24, 2026. All account data will be permanently deleted after cutoff with no recovery window. OpenAI cited unsustainable compute costs — reportedly $15M/day — and a strategic pivot toward coding tools and enterprise. If you haven't exported your content yet, go to sora.chatgpt.com/exports/me before the API cutoff.
The 2026 workflow meta has shifted decisively from single-shot prompting to systematic orchestration. Creators are now using specialized agent pipelines — a Director Agent drafts the shot list and camera angles, a Continuity Agent maintains wardrobe and lighting across shots. Front-loading critical requirements in prompts and using reinforcement keywords are now standard practice. The "Superagency" approach lets a single creator function as an entire production house.