Tasks and inputs
Genie 4 is listed for Multimodal, Text-to-video. It accepts text and produces video, audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Genie 4 is Google DeepMind's interactive world model with real-time native audio, available in Project Genie to eligible Google AI Ultra subscribers.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No API or per-use price; access comes with a Google AI Ultra subscription via Project Genie.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Most generative video ends when the clip does. Genie 4 does not hand you a clip at all: it builds a world you walk around in, live, in the browser, and keeps building it as you move.
Google DeepMind released it on 18 November 2026 inside Project Genie, the Google Labs experiment at labs.google/projectgenie. It is a world model, and Google describes it as its most realistic and interactive one so far.
Access is narrow, and it is worth being plain about that:
So, if you came here hoping to wire it into a product, the honest answer today is that you cannot. You can explore what it does, and judge whether it is worth tracking.
The big addition is sound. Genie 4 is the first Genie model to generate real-time native audio alongside its visuals, and that audio adapts to the scene, ranging from ambient soundscapes up to character dialogue. Visual fidelity and physics are also pitched as the best the series has offered.
You steer with text. A prompt starts the world, and further prompts mid-session alter it while you are inside, changing the weather, adding characters, or triggering actions.
Genie 3 has not gone away. It stays in Project Genie's model picker and picked up upgrades of its own: sessions now run three minutes per generation instead of sixty seconds, worlds can be chained into longer journeys, a 360-degree panorama can seed a world, and natural-language prompts can trigger in-world events.
Every piece of video and audio from Project Genie is marked with SynthID.
Google lists four open research problems: camera latency, intricate physical interactions, precision in the real-time audio, and keeping a scene consistent over long stretches. Expect objects you walked past to drift if you come back later.
No resolution, frame rate or session cap specific to Genie 4 has been published, so we have left those out rather than guess.
Pick a single scenario you care about, say a training environment, a game prototype or a pitch walkthrough. Then run it twice, once on each generation, using the same prompt. Watch three things: whether the audio actually matches what is on screen, how quickly the camera responds to input, and whether a landmark you leave is still the same when you return. That last test exposes the persistence limit faster than anything else.
If you need rendered clips for a pipeline rather than an explorable space, a conventional text-to-video model such as Veo 3.1 remains the practical choice, since it has developer access. The rest of Google's lineup sits on the Google provider page.
Compare the model on the work you need to complete.
Genie 4 is listed for Multimodal, Text-to-video. It accepts text and produces video, audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Apple
FastFrames-L is Apple's open 14B autoregressive two-step video diffusion model that generates 480p video at interactive latency, distilled from Wan2.1-14B.
Gemini Omni 1.2 Flash is Google's conversational video generation and editing model, making 3-15 second clips with native audio from 360p to 4K.
pika
Pika H3 Flex is Pika's open LoRA post-training of MiniMax H3: six-step sampling, up to 15 s clips with 32 kHz stereo audio at 480p or 768p, 24 fps.
Tencent
Hy-Nomas is Tencent Hunyuan's open interactive world model: it generates 720p video segment by segment, steered by text and camera controls.
heygen
HeyGen Video 1 generates 5 to 30 second MP4 clips with audio from text, images or reference media, up to 2K at 24 fps. Built on MiniMax H3.
Meta
Muse Video is Meta’s previewed video-generation model with native audio; a public Meta Model API route and price have not been established.