Google's Gemini Omni 1.2 Flash is now available on OpenRouter, extending clip length from 10 to 15 seconds and cutting per-second pricing by roughly 30%. It supports 360p to 4K output, native audio, and both 16:9 and 9:16 aspect ratios, while the same model also handles text, image, and video prompts via chat completions with a 131K context window.
Image-to-Video
图生视频
Technology that generates dynamic video sequences conditioned on static input images.
Curated Intelligence
Google released Gemini Omni 1.2 Flash, which debuts at #1 on Arena's Text-to-Video leaderboard with 1541 points, a 27-point gain over the prior version, and also ranks #1 in Image-to-Video and #3 in Video Edit. The update extends generation and editing to 15 seconds, improves temporal scene consistency and character speech, and cuts pricing 30% to $0.07 per second for 720p.
Video Rebirth released Hyperflow H3, an open-weight LoRA fine-tuned on MiniMax H3, ranking #5 in Image-to-Video and #3 in Video Editing benchmarks. It offers high-quality generation at less than half the price of competitor Dreamina Seedance 2.5, supporting 1088p resolution.
The Biological Computing Co. launched mischief-1 on OpenRouter, an image-to-video model built on four years of research using real brain cells grown on electrodes. Users provide a starting image and prompt to generate 3s or 5s clips at 480p or 720p resolution, with pricing starting at $0.005 per second.
Utopai Studios, an AI-native film production company, raised $60 million to reach a $1 billion valuation. Built on the custom video model Utopai X (based on MiniMax H3), which ranks second globally in blind evaluations for text-to-video and image-to-video generation, the company plans to release theatrical films by 2027 to validate its content production infrastructure.
Utopai Studios' video model Utopai X ranks #2 on the Artificial Analysis Image-to-Video (I2V) v2.1 benchmark, trailing only Dreamina Seedance 2.5. Post-trained on MiniMax H3, it is the only model in the top 2 for both Text-to-Video and I2V categories, excelling in physics simulation and multi-scene narratives. Access is currently restricted to the PAI platform with no public API.
Artificial Analysis has launched the AA-Video-I2V v2.1 and Silent benchmarks to evaluate image-to-video models. Based on over 58,000 human preference votes at up to 1080p resolution, it ranks models across 10 use cases, 10 capabilities, and diverse visual styles, emphasizing consistency and controllability.
fal has released FLUX 3 Max, a post-trained version of FLUX 3 that ranks #2 on AA-Video-I2V v1.1 (statistically tied for first) and #4 on AA-Video-T2V v2.1. The model generates 5-20 second, 720p videos with native audio from text or images, achieving high speed (under 12 seconds for a 10-second clip). It offers significantly lower pricing than the base FLUX 3 ($10.20/min vs $17.40/min) and excels in physics simulation and marketing use cases.
Shengshu Tech has released Vidu Q4 Preview, a new video generation model supporting up to 15 reference images and 3 audio clips for consistent visual and auditory features. The update significantly enhances dynamic camera movements, cuts, and complex shot transitions, offering up to 4K resolution with 10-bit depth and 16-second generation duration. With pricing starting at 0.09 RMB per second, it aims to reduce costs for multi-version testing in AI short dramas and advertising.
xAI released the low-cost video generation model Grok Imagine Video 1.5 Lite, ranking #17 on the Artificial Analysis Video Arena T2V v2.0 leaderboard and surpassing Google's Veo 3.1. Priced at $0.14 per second for 1080p (about one-third of Veo's cost) with a median generation time of 60.5 seconds, it occupies the quality-speed Pareto frontier. The model approaches top-tier performance in multi-scene narrative, lighting/materials, and text rendering, offering developers a high-value alternative.