Launch timeline
102 AI video launches from 29 companies since 1 June 2026, by company and month, then in full.
MethodEach entry links to the company's own announcement and, where one exists, an independent press report; where no announcement exists, a press report is used. Entries with one source are marked "single source". Dates are announcement dates. Weight: major model release, notable feature or version, or minor update such as pricing or an app. Official posts on X are shown as the original embed. LimitationCompanies come from a watchlist, so small or new companies can be missed until they are added.
Scroll the table sideways →
-
November 202617 releases
- 18 Nov
Claude connector · Synthesia
Synthesia brings its video agent into Claude via a connector. Users ask Claude to build a Synthesia video from a prompt, doc, deck or spreadsheet and refine it in chat; Synthesia says its Syren video agent uses Anthropic models.
- 18 Nov
Genie 4 in Project Genie · Google
Google rolls out Genie 4 to Google AI Ultra subscribers in Project Genie with mid-session world edits and a community gallery. Ultra subscribers in eligible regions can prompt changes while exploring generated worlds, with native audio and remixable shared worlds; AI Pro members can join a waitlist.
- 17 Nov
Phai-M1 · Phancy CN
Phancy (formerly 4Paradigm) launches Phai-M1, a video generation model for live-action style footage. Phai-M1 generates audio-synchronized 720p video of 15 seconds in about 3 minutes and was trained on Huawei Ascend chips; the company says weights will be opened on Hugging Face and ModelScope.
- 17 Nov
H3 Move in Suno Visuals · MiniMax CN
Suno launches Visuals music video feature running on H3 Move, a customized version of MiniMax H3. Suno Visuals generates clips of up to 15 seconds that follow a song's structure and lyrics with lip sync, using H3 Move, a fine-tuned and safety-filtered derivative of MiniMax H3.
- 17 Nov
Avatar 6 · HeyGen
HeyGen launches Avatar 6 talking avatar model with 2K output and no length limit. Avatar 6 generates a talking video of a person from a single photo, optionally with a short recording, at up to 2K (2880x1620) with no fixed duration limit and script-aware gestures; HeyGen reports 84.4% of side-by-side judgments found it indistinguishable from real footage.
- 17 Nov
Seedance 2.5 4K API · ByteDance CN
ByteDance opens the Seedance 2.5 4K API to developers on Volcano Engine and BytePlus at about 6.03 yuan per second. After an enterprise trial, the 4K version of Seedance 2.5 became available via API on Volcano Engine and BytePlus ModelArk, with 10-bit color and a separately billed low resolution draft workflow.
- 16 Nov
Short Drama Studio · Higgsfield
Higgsfield launches Short Drama Studio for turning a script into a multi-episode series. Uploads a script (PDF, DOCX, FDX, TXT or MD up to 10 MB), splits it into episodes, generates cast, locations and props, and renders episodes at 480p draft, 720p or 1080p in 15 styles; early preview on desktop and in Claude via Higgsfield MCP.
- 12 Nov
Seedance 2.5 flash · ByteDance CN
Volcano Engine launches Seedance 2.5 flash, a faster and lower cost tier of Seedance 2.5. Seedance 2.5 flash doubles generation speed versus Seedance 2.5 and costs about 34% less, roughly 0.99 yuan per second for 720p, supporting text, image and full multimodal reference to video, editing and extension at 480p and 720p.
- 12 Nov
Yijing video generation model · Baidu CN
Baidu Yijing launches an in-house video generation model priced from RMB 0.2 per second, plus a short drama studio. Announced at Baidu World 2026, the model offers a Fast version that generates 10 seconds of video in about 10 seconds, alongside a micro-expression model for digital humans and a Short Drama Studio with script and editing agents.
- 11 Nov
Project Live Set (sneak) · Adobe
Adobe previews Project Live Set, a real-time generative video editing prototype, at MAX 2026 Sneaks. The prototype lets editors change a subject, transform an environment, add a reference character or redirect the camera while a clip plays, with results shown in real time. It is a technology preview, not a shipping product.
- 10 Nov
Adobe and Google expanded partnership · Adobe
Adobe and Google deepen partnership to bring Google's latest image and video models to Firefly, Photoshop, Premiere and Express. Adobe customers can already use Gemini Omni, Veo and Nano Banana; Adobe and Google DeepMind will also collaborate on new AI capabilities, and Adobe tools come to Gemini and Google Workspace.
- 10 Nov
Firefly AI creative studio update (Adobe MAX 2026) · Adobe
Adobe expands Creative Agent across Firefly at MAX 2026, with a model selector and new video editing tools. Creative Agent can pick among Adobe and partner models from ByteDance, Google, Kling, OpenAI and others; Firefly adds storyboard-to-cut workflows, Firefly Graph (beta), and Premiere's Generative Media Tool expands with Generative Remove in beta.
- 9 Nov
Adobe and Disney AI alliance · Adobe
Adobe and Disney announce a multi-year alliance, with custom Firefly Foundry models trained on Disney properties. Adobe becomes Disney's Official AI Innovation Partner, giving early access to Firefly, Firefly Foundry, Premiere and Photoshop for Disney's production pipeline.
- 6 Nov
Higgsfield Miracle (private beta) · Higgsfield
Higgsfield previews Miracle, real-time layer-based video editing, in private beta. Miracle splits video into layers so a described change, such as recoloring one object, applies while the clip plays and the rest of the frame stays unchanged; waitlist only, coming first to Higgsfield MCP.
- 6 Nov
Micro Drama Ads · Higgsfield
Higgsfield adds agent-built video ads, including micro drama story ads, to Ads Studio. An agent in chat builds story ads with a character, conflict and product at the turning point across 8 styles, keeping the brand kit product and avatar consistent across ads; no length limit, 9:16, also available via Higgsfield MCP.
- 5 Nov
Gemini Omni 1.2 Flash · Google
Google releases Gemini Omni 1.2 Flash with native 15-second clips and a 30% price cut, generally available on Gemini API and Google Cloud. Priced at $0.07 per second for 720p, up to 15-second clips (from 10), stronger reference adherence, multi-turn editing and 360p/720p drafts upscalable to 1080p or 4K. Google Vids users get three free 15-second videos per month.
- 3 Nov
ChatGPT plugin · Synthesia
Synthesia launches a ChatGPT plugin for creating avatar videos in chat. Users describe a video or share a file in ChatGPT and create and refine finished Synthesia videos without leaving the chat; available on all Synthesia plans.
- 18 Nov
-
October 202620 releases
- 29 Oct
Pika H3 Flex · Pika
Pika releases H3 Flex, a post-trained MiniMax H3 video model with open weights. H3 Flex generates up to 15 seconds of video with stereo audio natively at 480p or 768p, takes up to 12 references, offers five audio stems via API and hosted upscaling to 4K; weights and inference code are on Hugging Face and GitHub.
- 27 Oct
Genie 4 (announced) · Google
Google DeepMind announces Genie 4 world model with native audio and real-time steering by text prompt. Genie 4 builds on Gemini Omni for real-world appearance and lets users change camera, weather or characters mid-simulation, with generated ambient sound, music and dialogue. Initially limited to trusted testers.
- 26 Oct
App Studio · Runway
Runway launches App Studio for building custom creative apps from plain language. App Studio builds hosted apps with their own interface that generate images, video and audio using a team's Brand Kits and Workflows, on paid plans.
- 26 Oct
Higgsfield Playable · Higgsfield
Higgsfield launches Playable for generating browser games and playable ads. Playable builds a mobile game from a text description or gameplay video across 8 genres in 2D or 3D; the Enterprise Playable Ads tier exports a single HTML file up to 5 MB for ad networks such as AppLovin and Moloco.
- 26 Oct
Jimeng Short Drama Studio · ByteDance CN
ByteDance's Jimeng launches Short Drama Studio, an end-to-end workspace for AI short drama production. The studio combines script development, character and scene management, storyboarding, Seedance video generation and post-production in one project space, alongside new in-house models DreamScript (scripts), Anycook (images) and DreamShot (storyboards).
- 23 Oct
Hy-Nomas · Tencent CN
Tencent open-sources Hy-Nomas, an interactive video world model with text and camera control. Hy-Nomas is an 8B-parameter model that generates explorable 720p worlds for up to 60 seconds, with multi-turn text prompts, camera trajectories and keyboard control, released under Apache 2.0.
- 23 Oct
Seedance 2.5 4K · ByteDance CN
Volcano Engine releases Seedance 2.5 4K for film and TV production, initially for selected enterprise customers. Seedance 2.5 4K adds 4K output with 10-bit color and a Draft mode that renders lower cost 480p previews before the final 4K pass; it opened first as a trial for selected enterprise and post-production partners.
- 22 Oct
Content Factory · Higgsfield
Higgsfield launches Content Factory for turning a topic into a ready-to-post video series. Content Factory writes the script, builds characters and scenes with voiceover and edits the clip; clips run 15 seconds to 5 minutes, 1 to 4 episodes per run, with 50+ format presets. Also usable inside Claude via Higgsfield MCP.
- 22 Oct
Veo 3.1 preview models retired in Gemini API · Google
Google shuts down Veo 3.1 preview models in the Gemini API, naming Gemini Omni 1.2 Flash as the replacement. veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview were shut down on Oct 22, 2026; Veo 2.0 and 3.0 GA models were shut down in the Gemini API on Jun 30, 2026. Veo GA models remain on Gemini Enterprise Agent Platform.
- 21 Oct
Kling 4.0 API · Kuaishou CN
Kling AI opens the Kling 4.0 API to enterprise customers and developers. Kling announced on its official Weibo that Kling 4.0 API service went live with applications open, rolling out gradually to global enterprise customers and developers; Perfect Corp's YouCam was among the first integrators.
- 19 Oct
FLUX 3 Max · fal
fal launches FLUX 3 Max, a faster FLUX 3 video model post-trained with Black Forest Labs. fal reports a 10-second video in 10.95 seconds and a 1,139 Elo on the Artificial Analysis image-to-video arena.
- 19 Oct
Marketing Studio and rebuilt Supercomputer · Higgsfield
Higgsfield launches new Marketing Studio and rebuilt Supercomputer for campaign production. Marketing Studio turns a product URL into a brand kit and campaigns from 1,500+ templates; the rebuilt Supercomputer routes work across OpenAI, Anthropic, Google, Kling and Seedance models with AI Employees, shared skills, scheduled tasks and 30+ connectors.
- 15 Oct
Krea 3 · Krea
Krea releases Krea 3, an in-house image editing model built for speed and cost. Krea 3 became the default image model in the Krea app; an Ultra tier and 4K API output followed.
- 15 Oct
Kling 4.0 · Kuaishou CN
Kling AI launches Kling 4.0 to all users with 30 second native generation, 10-bit HDR and up to 10 keyframes. Kling 4.0 went fully live on klingai.com with native 30 second clips, extension up to two minutes, 4K and 1080p output with up to 10-bit HDR, 21:9 framing, up to 15 Omni Reference inputs, synchronized audio and 8,000 token prompts; it is also available through Kling MCP and CLI, with the API announced as coming soon.
- 8 Oct
Runway in Claude Motion · Runway
Runway is a launch partner for Anthropic's Claude Motion. Animations made in Claude Motion can be exported to Runway for added generated footage and editing.
- 7 Oct
Vidu Q4 Preview · Shengshu CN
Shengshu releases Vidu Q4 Preview, its next flagship video model, with 4K output and launch pricing from $0.014 per second. Vidu Q4 Preview supports up to 15 image references and 3 audio references, 2K or 4K output with 10-bit color, and focuses on character performance and voice consistency; China pricing starts at RMB 0.09 per second.
- 2 Oct
AI Influencer · Higgsfield
Higgsfield launches AI Influencer tool for building custom characters and placing them in videos. Users build a character from scratch, a reference or their own photo with a 19-category Character Builder, then insert it into existing videos via Genjutsu. Semafor reported the designs are deliberately non-photorealistic.
- 1 Oct
Griffin · Tavus
Tavus introduces Griffin, a real-time face-to-face human interaction model. Griffin powers real-time video conversations with a lifelike avatar; Tavus says 48% of participants in a one-minute video call test thought the research-preview Griffin-Lite was human. Early access is limited to select developers.
- 1 Oct
Synthesia Sessions · Synthesia
Synthesia launches Sessions for two-way conversations with interactive AI avatars. Sessions launched with Roleplay and Survey Sessions for training, coaching and research, with live sessions announced as a third type.
- 1 Oct
Comfy Agent · Comfy Org
Comfy Org launches Comfy Agent, which plans, builds and runs ComfyUI workflows from text. Users describe the result inside ComfyUI and the agent assembles and executes the node workflow.
- 29 Oct
-
September 202622 releases
- 30 Sep
All-new PAI (PAI 3.0) and PAI Code · Utopai Studios
Utopai Studios launches all-new PAI production platform with self-serve plans and PAI Code. The rebuilt PAI keeps project memory of story, characters and style, with a production library, versioned approvals and a prompt ledger; self-serve plans run from $15 to $379 a month, and PAI Code brings PAI skills into Claude Code and Codex.
- 30 Sep
Utopai X · Utopai Studios
Utopai Studios releases Utopai X video model, ranked No. 2 on Artificial Analysis text-to-video with audio. Utopai X is a text-to-video model with audio that Utopai says is post-trained on MiniMax H3; it debuted at 1,150 Elo, No. 2 on the Artificial Analysis leaderboard, and is available inside PAI rather than through a public API.
- 30 Sep
Runway Ads · Runway
Runway announces Runway Ads to generate, publish and optimize ad creative on Meta, Google and TikTok. Runway Ads connects to ad accounts and a brand kit, publishes approved variants and uses performance data to produce new rounds; it became generally available on all paid plans on Oct 27.
- 30 Sep
HeyGen Video 1 · HeyGen
HeyGen releases HeyGen Video 1, its first general-purpose video model. Generates picture, dialogue and sound effects in one pass from text, image or reference inputs; the API launched with 5 to 15 second clips at 480p or 768p, later extended to 1080p, 2K and 30 seconds, priced from $0.02 per second. It reached No. 1 on Artificial Analysis' silent text-to-video board in October.
- 28 Sep
Express-3 · Synthesia
Synthesia releases Express-3 avatar model to all users. Synthesia describes Express-3 as its best avatar model, faster, sharper and more expressive, with styling options.
- 28 Sep
Kling 4.0 (announced) and Kling 4.0 Flash · Kuaishou CN
Kling AI announces Kling 4.0 for October release and opens Kling 4.0 Flash to a small group of users. Kling 4.0 was announced with up to 30 second generation, 4K and 1080p 10-bit HDR output, up to 10 keyframes and 8,000 token prompts; the lighter Kling 4.0 Flash (720p, 3 to 20 seconds) opened first to Ultra annual subscribers in a closed beta.
- 24 Sep
Sora 2 API shutdown · OpenAI
OpenAI removes the Sora 2 Videos API, completing the Sora wind-down announced in March. The Videos API and sora-2 and sora-2-pro model snapshots reached end of life on Sep 24, 2026 with no one-to-one replacement. The Sora app and web experience had closed on Apr 26, 2026.
- 23 Sep
Muse Realtime Avatar · Meta
Meta introduces Muse Realtime Avatar, generating interactive talking avatar video in real time for its Muse agent. Turns an image into an avatar with facial, hand and body motion at 448x768 and 25 fps, about 870 ms response latency. Video chat with Muse avatars rolled out to US users on Nov 5, 2026.
- 23 Sep
FLUX 3 Action · Black Forest Labs
Black Forest Labs releases FLUX 3 Action, a 7B open-weights world action model for robots. The model applies FLUX 3 video pretraining to robot control.
- 23 Sep
Free Gemini Omni 1.1 video in Google Vids · Google
Google opens free Gemini Omni 1.1 Flash video generation in Google Vids to anyone with a Google account. Personal Google and Workspace accounts can generate 1080p AI video scenes at no cost via vids.new on desktop, with control over scene duration and transitions.
- 23 Sep
Comfy Router · Comfy Org
Comfy Org introduces Comfy Router, one API for frontier image and video models. Developers call multiple partner media models through one API and choose where each job runs.
- 22 Sep
PixVerse R2 · AISphere CN
PixVerse introduces R2, an upgrade to its real-time world model with session memory and multimodal input. R2 stays coherent over longer sessions, remembers events within a session, and accepts text, references, audio and action controls inside the same running world, following R1 from January 2026.
- 17 Sep
New Pika platform · Pika
Pika relaunches as a unified creative platform for video, image and sound. The new Pika puts third-party and Pika models behind focused creative apps, including shot-list prompting, relighting and character swap tools.
- 16 Sep
Higgsfield API · Higgsfield
Higgsfield launches a standalone pay-per-generation API covering 50+ models. One key opens 50+ image, video and audio models with Python and TypeScript SDKs; billing is per successful generation and separate from higgsfield.ai plans and credits.
- 15 Sep
Vidu S2 · Shengshu CN
Shengshu unveils Vidu S2, adding real-time avatar interaction at 720p and real-time video stream editing. Vidu S2 has two models: S2-Avatar raises real-time output from 540p to 720p and accepts new reference images mid-generation, and S2-Editing changes style, outfit, subject or background of an incoming video stream in real time.
- 8 Sep
Runway for Premiere Pro and After Effects · Runway
Runway launches plugins for Adobe Premiere Pro and After Effects. A Runway panel inside both apps generates images and video, restyles timeline footage with Aleph 2 and places results in the sequence; free download for macOS and Windows using existing paid plans.
- 8 Sep
Generative Media Tool in Premiere · Adobe
Adobe Premiere adds a Generative Media Tool for creating video and sound effects directly in the timeline. Generate Video fills timeline gaps using reference frames from the project, with a choice of Firefly or partner models including Google Veo, Kling, Runway and Luma; Generate Soundscape (beta) analyzes up to 15 seconds of video.
- 4 Sep
3D Jutsu · Higgsfield
Higgsfield launches 3D Jutsu, a 3D scene workspace that renders to video. 3D Jutsu turns a prompt or reference into an editable 3D scene with objects, lighting, cameras and animation, then generates finished video; it works with Supercomputer and models including GPT-6 Astra.
- 3 Sep
GWM Worlds 2 · Runway
Runway shares GWM Worlds 2, a real-time interactive world model with audio. Built on Runway's audio-video generation model, it lets users define an environment and steer it with text actions and continuous camera motion.
- 2 Sep
Arabic interface · Luma AI
Luma AI says Luma Agents will get a native Arabic-language interface for Saudi and Arabic-speaking creators. The fully Arabic interface is scheduled for later in 2026, tied to Luma's partnership with HUMAIN.
- 1 Sep
H3 Max · fal
fal launches H3 Max, a video model post-trained from the open-weights MiniMax H3. H3 Max generates 768p text-to-video and image-to-video with audio, about 3 seconds of wall time for a 5 second clip per fal, at $0.08 per second after a $0.04 launch promotion; available on the fal API and Playground.
- 1 Sep
Genjutsu · Higgsfield
Higgsfield launches Genjutsu for swapping objects, locations and characters in existing video. Genjutsu changes an object, location, product or character in a video while leaving the rest untouched, with Motion Transfer and Object Swap modes; inputs are one video up to 30 seconds and up to 30 reference photos, output up to 1080p. A Restyle mode followed on September 30 and a cheaper Genjutsu Lite on October 19.
- 30 Sep
-
August 202616 releases
- 31 Aug
Solaris interface world model · Runway
Runway introduces Solaris, a real-time model that generates software interfaces frame by frame. Solaris is the first of what Runway calls Interface World Models, rendering app and website interfaces as live video instead of code.
- 27 Aug
Gemini Omni 1.1 Flash · Google
Google releases Gemini Omni 1.1 Flash with scene extension to 40 seconds, first and last frame control and 4K upscaling. Adds 360p draft previews, start and end frame interpolation, scene extension up to 40 seconds and 1080p or 4K export. Available in Google Flow, AI Studio and Gemini Enterprise Agent Platform.
- 24 Aug
Wan3.0 · Alibaba CN
Alibaba launches Wan3.0 video model commercially with 30 second generation and document to video input. Wan3.0 generates up to 30 second clips and accepts DOC, XLS, PPT, PDF and Markdown files, unifying text, image and reference to video plus editing in one model; Alibaba Cloud offered 30% off API pricing through September 23. As of November 2026 no Wan3.0 weights appear on the Wan-AI Hugging Face page.
- 20 Aug
Gemini Omni Flash in Firefly · Adobe
Adobe adds Google's Gemini Omni Flash video model to Firefly and makes music, speech and sound effect generation generally available. Gemini Omni Flash joins partner models from Google, Kling AI, Luma AI, OpenAI and Runway in Firefly; the same update added Generate Music, Generate Speech and Generate Sound Effects and a free Firefly AI Assistant tier.
- 17 Aug
ByteDance and MPA copyright safeguards agreement · ByteDance CN
ByteDance and the Motion Picture Association sign an agreement on copyright safeguards for Seedance and Seedream. The pact strengthens IP protections on ByteDance's AI video and image generation models after Hollywood studios objected to Seedance 2.0 outputs earlier in 2026.
- 14 Aug
Pika Audio models · Pika
Pika releases four audio models: Soundtrack, Music, SFX and Speech. Soundtrack generates synchronized sound for video, Music writes full songs, SFX makes sound effects and Speech is a 3B text-to-speech model with voice cloning; Pika says prices are up to 20x lower than comparable models.
- 13 Aug
Dumbstruck partnership · Luma AI
Luma and Dumbstruck launch Creative Intelligence, a decision layer for ad creative. The joint product pairs Luma generation with Dumbstruck audience testing data for advertising teams.
- 12 Aug
Cinema Studio 4.0 · Higgsfield
Higgsfield releases Cinema Studio 4.0 with 30-second generations and up to 50 references. Version 4.0 extends generations to 30 seconds and up to 50 references, with 30+ camera presets, a rebuilt 5-lens system, 50+ color palette presets and an Emotion Wheel for character performance. Higgsfield issued a formal press release with a Director's Panel on October 28.
- 11 Aug
Luma Scenes · Luma AI
Luma launches Scenes, a storyboard-first tool that renders video only after keyframes are approved. Scenes uses Uni-1 to build an editable keyframe storyboard from a brief or brand assets, then renders with Ray 3.2 or Seedance 2 in 9:16 or 16:9.
- 11 Aug
LTX-2.5 · LTX (Lightricks)
LTX releases LTX-2.5 open weights audio-video model with native multishot. LTX-2.5 adds native multishot, a custom Gemma 4 12B text encoder, EXR workflows and a pretrained checkpoint for fine-tuning, with day-one ComfyUI support; VentureBeat reports a 10-second clip from an image in 6.8 seconds on Nvidia hardware.
- 7 Aug
Wan-Animate-2 · Alibaba CN
Alibaba open sources Wan-Animate-2 character animation model weights under Apache 2.0. Wan-Animate-2 drives character animation directly from a driving video without skeletal pose extraction, adds text driven viewpoint control, and ships Base and Distillation weights plus a real time Lite variant.
- 6 Aug
Wan3.0 public beta · Alibaba CN
Alibaba opens Wan3.0 video model to public beta with 30 second clips and document inputs. Wan3.0 doubles maximum clip length to 30 seconds from Wan2.7-Video's 15 and accepts text, images, video, audio, web pages, PDFs and slides as inputs; testers can apply via Model Studio and Qwen Cloud.
- 4 Aug
MAGI-2 Preview · Sand.ai CN
Sand.ai open-sources MAGI-2 Preview, a 114B-parameter mixture-of-experts audio-video generation model. MAGI-2 Preview activates about 6B parameters per token and generates video with audio through a preview stage plus a refiner; weights are on Hugging Face.
- 4 Aug
Pika API Club · Pika
Pika launches API Club, a $10 per month membership for discounted access to 100+ models. Members pay $10 a month, returned as $10 in API credits, for access to video, image, audio and LLM models at near-wholesale prices. Announced via a Business Wire release on Aug 4 (blog post dated Aug 5).
- 4 Aug
FLUX 3 Video general availability · Black Forest Labs
Black Forest Labs makes FLUX 3 Video generally available at 720p and 1080p. FLUX 3 Video generates up to 20 seconds with native audio, supports text, image and keyframe input, video continuation, multiple shots and multilingual lip-synced dialogue.
- 3 Aug
MiniMax H3 open weights · MiniMax CN
MiniMax publishes MiniMax H3 video model weights on Hugging Face under a community license. The weights for H3 (text, image, video and audio to video with synchronized audio) went live on Hugging Face on August 3, with chip vendors and developer platforms announcing same-day support.
- 31 Aug
-
July 202610 releases
- 31 Jul
Imagine Video 1.5 with References · xAI
xAI adds text-to-video, native 1080p and image and voice references to Grok Imagine Video 1.5. Up to seven reference images per generation plus voice references for consistent characters; text-to-video and 1080p are live on grok.com, iOS, Android and the xAI API, with voice references in the API on request.
- 31 Jul
MiniMax H3 · MiniMax CN
MiniMax releases H3 (Hailuo 3.0), a multimodal video model with native stereo audio, up to 15 seconds at 2K. H3 takes text, image, video and audio as context and generates video with native stereo sound at up to 2K resolution and 15 seconds; MiniMax says its 2K per-second price is under a third of mainstream models.
- 31 Jul
Seedance 2.5 · ByteDance CN
ByteDance releases Seedance 2.5, debuting on Jimeng and Dreamina with 30 second audio-video generation. Seedance 2.5 generates 30 second clips with joint audio in a single run, supports two further extensions and up to 30 images, 10 videos and 10 audio clips as references; it launched first on Jimeng/Dreamina and Doubao Pro, and Jimeng added Maya and Blender plugins and a smart editing mode alongside it.
- 28 Jul
Mirage Avatar X · Mirage
Mirage releases Avatar X avatar model, now powering all AI avatars in Captions. Avatar X generates the whole avatar performance, expressions, voice and motion, in one pass and now powers Captions' AI avatars and AI twins in horizontal or vertical formats.
- 23 Jul
Runway Dev and Media Router · Runway
Runway launches Runway Dev developer platform and Media Router for automatic model selection. Runway Dev bundles Runway and third-party models, recipes, workflows and Characters behind one API; Media Router picks a video, image or audio model per request based on cost, quality and latency settings.
- 23 Jul
FLUX 3 · Black Forest Labs
Black Forest Labs unveils FLUX 3, its first model to generate video with audio. FLUX 3 is trained jointly on video, images and audio and generates clips with native audio up to 20 seconds; it launched in limited release.
- 16 Jul
Gemini Omni and personal avatars in Google Vids · Google
Google Vids adds Gemini Omni video generation and editing plus personal avatars built from a selfie and voice recording. Users can generate and edit clips with text prompts and image references inside Vids, and create a digital avatar of themselves. AI clips carry a digital watermark.
- 7 Jul
Muse Video (preview) · Meta
Meta previews Muse Video, its video model with native audio, alongside the launch of Muse Image. Meta said Muse Video ranks third in Arena text-to-video human preference and is coming soon to creators and Meta AI; it has not given a public release date. Muse Image launched the same day in Meta AI, Instagram Stories and WhatsApp.
- 7 Jul
Higgsfield Apps · Higgsfield
Higgsfield launches Apps: Supercomputer builds generative image, video and 3D apps. Supercomputer builds and publishes full generative apps that call Higgsfield models with no API keys, up to 20 apps a day; app users generate on their own Higgsfield credits. Also available in Claude and Cursor via Higgsfield MCP, launched with a $100,000 app contest.
- 3 Jul
Vidu S1 · Shengshu CN
Shengshu unveils Vidu S1, a real-time interactive video model with voice-controlled characters. Vidu S1 generates 540p video at 25 fps (up to 42 fps) for live voice-driven conversations with characters created from a single image, and runs on consumer-grade GPUs.
- 31 Jul
-
June 202617 releases
- 30 Jun
Gemini Omni Flash (API public preview) · Google
Google brings Gemini Omni Flash video model to the Gemini API and AI Studio in public preview, alongside Nano Banana 2 Lite. Gemini Omni Flash, first shown at Google I/O in May, supports video generation and conversational editing from text, image and video inputs. Priced at $0.10 per second of output, the same as Veo 3.1 Fast.
- 29 Jun
MIXI partnership · Runway
Runway and Japan's MIXI announce strategic enterprise partnership. MIXI has deployed Runway across creative operations and plans joint work on gaming, animation and interactive uses.
- 29 Jun
Comfy MCP · Comfy Org
Comfy Org releases Comfy MCP to connect AI agents to the ComfyUI ecosystem. Claude, Codex, Cursor and other MCP clients can build and run ComfyUI image and video workflows; a local open-source version followed on Aug 18.
- 26 Jun
Unlimited Models Marketplace · Higgsfield
Higgsfield introduces Unlimited Models Marketplace with fixed-window access to top models. Users buy unlimited generations on a single model (Nano Banana Pro, Seedance 2.0, GPT Image 2, Kling 3.0 and others) for 1, 7 or 14 days instead of paying credits per generation; MCP, CLI, Canvas and Supercomputer still use credits. Extended to Team and Scale plans on July 4.
- 25 Jun
Agent 2.0 · Runway
Runway launches Agent 2.0, an upgraded agent for producing ads and marketing campaigns. Agent 2.0 creates campaigns, analyzes performance data to improve creative and adapts output across platforms, formats and markets.
- 23 Jun
Seedance 2.5 (announced) · ByteDance CN
ByteDance announces Seedance 2.5 at Volcano Engine FORCE conference with 30 second clips and 50 reference inputs. Previewed in Beijing, Seedance 2.5 targets native 30 second single pass video, native 4K and up to 50 multimodal reference inputs per generation; Volcano Engine's model release log lists doubao-seedance-2-5 under June 2026.
- 22 Jun
HappyHorse 1.1 · Alibaba CN
Alibaba releases HappyHorse 1.1 video model with stronger motion, consistency and audio sync. The upgrade from Alibaba's ATH unit improves reference to video character consistency, motion, instruction following and audio-visual synchronization, and is available on happyhorse.com and via API on Model Studio.
- 19 Jun
Supercomputer 2.0 · Higgsfield
Higgsfield launches Supercomputer 2.0 agent layer for enterprise marketing on NVIDIA. Higgsfield says Supercomputer 2.0 can run campaigns from ideation to live posting; the same day it shipped AI Employees 2.0 with memory and connectors that persist across sessions.
- 18 Jun
Firefly AI Assistant and upgraded creative AI studio · Adobe
Adobe adds agentic Firefly AI Assistant features including product-photo-to-video and video generation from storyboards. Firefly AI Assistant can turn product photos into short videos, build storyboards and generate video from them, and assemble footage with Quick Cut; Adobe also previewed a redesigned Firefly studio with Projects.
- 17 Jun
LTX Trainer unified framework · LTX (Lightricks)
LTX releases a major LTX Trainer update with unified video, audio and cross-modal training. The open-source framework now covers LoRA, IC-LoRA and full fine-tunes across T2V, I2V, audio, A2V and V2A modes, plus agentic config generation.
- 17 Jun
Kling 3.0 Turbo · Kuaishou CN
Kling AI releases Kling 3.0 Turbo, a faster and cheaper video model with native audio. Turbo generates 1 to 15 second clips from text or multiple reference images with native audio; Kling's developer pricing lists it at $0.112 per second for 720p and $0.14 per second for 1080p.
- 16 Jun
Grok Imagine Video 1.5 · xAI
xAI makes Grok Imagine Video 1.5 generally available in its API and ships Video 1.5 Fast in Grok apps. Improves motion, physics and same-pass audio and speech. Video 1.5 Fast generates 6-second 720p clips in about 25 seconds, down from 40+ seconds; Grok Imagine also adds Projects, parallel agents and library search.
- 15 Jun
MiniMax Hub · MiniMax CN
MiniMax launches Hub, an agent-driven video creation workspace, at the Shanghai International Film Festival. Hub combines image, video, voiceover, music and editing in one platform, with an AI agent that breaks a brief into tasks, picks models and checks quality while the user reviews at key points.
- 11 Jun
Lionsgate partnership expansion · Runway
Runway and Lionsgate expand partnership; Lionsgate takes equity stake in Runway. The companies announced a joint development program to create new IP, with Lionsgate drawing on its catalog for an AI short-form series.
- 9 Jun
Ray3.2 · Luma AI
Luma AI releases Ray3.2 video model with frame-level control and its first full Ray API. Ray3.2 adds multi-keyframe direction (up to 16 keyframes per clip), Modify Video V2 for up to 20 seconds at 1080p, HDR with EXR output, and exposes the full Ray control surface through an API.
- 3 Jun
Grok Imagine Video 1.5 Preview · xAI
xAI releases grok-imagine-video-1.5-preview image-to-video model in the xAI API. Animates a single still image into video with camera moves and sound, at up to 720p; offered as an API preview.
- 2 Jun
PAI 2.0 · Utopai Studios
Utopai Studios launches PAI 2.0 with a creative agent and 4K output. PAI 2.0 adds character consistency, a story agent, a canvas workspace, voice input and 4K output, less than three months after PAI's public release.
- 30 Jun