MiniMax Hailuo 02 Tops Benchmarks, Midjourney Debuts V1

MiniMax Hailuo 02 Sets New Benchmark for AI Video Generation

This week saw a seismic shift in the competitive AI video space as MiniMax officially launched Hailuo 02, its latest video generation model. Released on June 18, 2025, Hailuo 02 has immediately taken the spotlight for its performance, efficiency, and ability to generate visually compelling, dynamic clips at unprecedented cost levels.

• Advanced Physical Realism: Hailuo 02 rose to global viral prominence thanks to its Olympic diving cat videos, which captivated social media with nuanced fur movement, water splash effects, and realistic motion captured frame by frame.

• Innovative Architecture: Powering Hailuo 02 is the Noise-aware Compute Redistribution (NCR) architecture, which increases training and inference speed by a reported 2.5x compared to previous MiniMax models.

• Flexible Output: The model comes in three distinct versions, 768p for 6 seconds, 768p for 10 seconds, and 1080p for 6 seconds, allowing for tailored use cases from quick creative sketches to commercial-grade clips.

• User Accessibility and Cost: Hailuo 02 is designed to be broadly accessible: users do not need enterprise ties or closed beta invites. This simplicity, combined with efficient cloud cost performance, positions Hailuo 02 as a top contender for both professional and hobbyist creators worldwide.

The benchmark response has been significant: according to early user reviews and influencer benchmarks, Hailuo 02 surpassed Google’s Veo 3 in several user benchmarks, particularly excelling in motion continuity, subject adherence, and video cost per second. Advanced features, including enhanced temporal synchronization and nuanced scene understanding, have led to buzz beyond China, catalyzing new creative trends. Many experts point to MiniMax’s agility in fusing cutting-edge academic research with a product-first deployment approach.

On the horizon for Hailuo 02 are upcoming features such as audio synchronization, extended lighting controls, and longer-duration clip generation, promising to push the boundaries of AI-generated cinematic content even further.

Midjourney V1 Video Model Launches to Public

Also this week, leading visual creativity platform Midjourney launched the V1 model to general users on June 18, 2025, marking its formal expansion beyond still image artistry into the realm of dynamic video. V1 represents a leap for the brand, leveraging its reputation for artistic visual output by translating it into moving scenes.

• Image-to-Video Transformation: Users upload either a Midjourney-generated image or an external image and create a set of four five-second video clips. The process is entirely Discord-based, maintaining Midjourney’s community aesthetic.

• Creative Motion Control: A spectrum of control is offered, from "automatic animation" with randomized effects to detailed manual descriptions, plus "low motion" and "high motion" settings for adjusting camera and character activity.

• Longer Clips, More Iteration: Videos can be extended by four seconds up to four times, reaching a maximum of 21 seconds. Higher tier subscription plans offer broad access to relax-mode generations for power users.

• Priced for Everyone: Entry at $10/month, with resource usage balanced so casual and power creators can both experiment without friction. V1 also underscores Midjourney’s ambitions to make video generation as approachable as image prompting.

Initial user feedback highlights smooth transitions, photorealistic interpolations, and subject stability, distinct Midjourney hallmarks, now in motion. While the first model prioritizes ease and accessibility over raw video realism, V1’s roadmap shares a clear ambition: paving the way towards true real-time multi-modal creative tools for artists everywhere.

China Races Ahead in the AI Video Arms Race

With this pair of headline launches, the generative video space is now dominated by a cross-Pacific rivalry. On the Chinese front, MiniMax’s public launch of Hailuo 02 has set a new technical and creative bar. The U.S. and international landscape, meanwhile, sees Midjourney moving rapidly to democratize creative video workflows for a global creator base. The rapid escalation of video output realism, efficiency, and accessibility underscores the exponential pace of progress, and signals broader market dynamics ahead.

No New Open Source Video Model Releases This Week

While open source remains a vital undercurrent in AI video, this week saw no publicly released text-to-video or image-to-video models on platforms like Hugging Face. The latest impactful open source releases remain earlier in 2025, including the compact SmolVLM family for mobile video and image processing. Researchers and developers seeking fully transparent codebases will need to wait a bit longer for the next wave, although the rapid arms race at the commercial level may soon inspire an open source response.

Honorable Mentions

• OpenAI Sora: Now available for ChatGPT Plus and Pro users, allowing the creation of 20-second, 1080p videos. Ongoing refinement and content controls keep Sora in the discussion, though no major update landed this week.

• Runway Gen-4: Continues to impress with cross-scene subject coherence and creative visual prompt interpretation, although its main release came earlier this spring.

The video generation field continues its relentless acceleration, with new commercial titans emerging, creator communities rapidly scaling up, and a keen sense of anticipation for open systems to match the commercial arms race in creativity, efficiency, and raw power.