Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
Runway's Gen-4.5, built with NVIDIA using Autoregressive-to-Diffusion techniques, now sits at the top of the Artificial Analysis text-to-video leaderboard with an Elo score of 1,247. It supports image-to-video, keyframes, video-to-video, and speed control. The separately-released Runway Agent (May 13) lets you describe a concept, refine it conversationally, and generate entirely within one session.
NVIDIA's 2.6B-parameter SANA-WM dropped May 16 and it's a beast on a budget: one image plus text plus a camera trajectory generates up to 60 seconds of 720p with 6-DoF camera control. A Gated DeltaNet recurrence replaces softmax attention, keeping memory constant regardless of clip length. Distilled variant clocks 60 seconds of 720p in roughly 34 seconds on a single 5090.
A community-released modality-wise reinforcement learning LoRA for LTX-2.3's Twin-DiT architecture is showing substantial gains across AV-Quality, AV-Consistency, and AV-Synchrony — with a reported 52% reduction in sync errors. It's live on Hugging Face as LTX-2.3-OmniNFT-RL-Lora_bf16 and loads in Kijai's LTX2.3_comfy node set. Community calling it one of the most important upgrades currently available for LTX 2.3 workflows.
TikTok is now using C2PA Content Credentials to auto-detect synthetic media — even without creator self-disclosure. Unlabeled AI video flagged by the system results in a 14-day Creator Fund earnings freeze, even if you retroactively add the label. Deepfakes of real private individuals are banned outright, label or no label. This is the most aggressive enforcement posture any major platform has taken so far.
Google dropped Veo 3.1 Lite this month, targeting high-volume developer apps. It runs at the same speed as Veo 3.1 Fast but costs less than half as much. The parent Veo 3.1 also got an "Ingredients to Video" update adding native vertical output for mobile-first short-form, 4K upscaling, and identity consistency across scene changes. Available now through the Gemini API and Vertex AI.
The May 20 release of ComfyUI 0.22.0 packs meaningful additions: Stable Audio 3.0 support, MoGe geometry models, HiDream-O1 with area conditioning, downscaled IC-LoRA for LTX 2.3, and a new downscale_ratio_temporal control for video temporal pacing. Peak VRAM is reduced when guide_mask is active in LTX 2.3 workflows. RIFE and FILM frame interpolation nodes and SAM 3.1 segmentation also shipped.
Kling 3.0 is currently sitting atop the community arena with a score of 2,127. Its AI Director feature generates six shots per clip with autonomous camera motion and scene continuity. Omni Native Audio adds multilingual lip-sync in Japanese, Korean, Spanish, and English with environmental soundscapes baked in. The Motion Brush gives per-frame path drawing for fine motion control. Native 4K at 60fps, up to 15 seconds per clip.
Wan 2.7 from Alibaba's Tongyi Lab continues to expand its reach with a four-model suite covering text-to-video, image-to-video, reference-to-video, and video editing. The 9-grid multi-angle input — a 3x3 arrangement of reference images — enables multi-angle consistency across clips. Video-to-video editing accepts a clip plus a plain-text instruction and returns the edited output. At $0.10 per second, it's the most cost-competitive high-quality API in the space right now.
Announced at Google I/O May 19, Gemini Omni combines Veo, Genie, and Gemini Nano under one multimodal roof — text, image, audio, and video in, video out — with a "world model" framing that simulates physical logic across edits and maintains consistent characters and backgrounds. It's rolling into Flow and YouTube Shorts Remix, and is free in the YouTube Create app. This one has structural implications for short-form platform distribution.
OpenAI quietly killed the Sora web and mobile app on April 26, 2026. The Sora 2 API is still active — supporting up to 20-second generations, 1080p on the sora-2-pro endpoint at $0.70 per second, reusable character references, and video extensions. The API deprecates September 24. If you're building on Sora 2 API, start planning your migration now.
