New models and features in AI image, video, voice and music generation, from Midjourney, Sora, Veo, Suno and more.
Quiver just launched Arrow 1.1 and Arrow 1.1 Max, upgrading its text-to-SVG vector models. Instead of approximating pixels, the new models construct cleaner SVGs using basic geometric primitives and editable text, cutting generation costs by 33.3% and vectorization costs by 50%. The Max edition targets pixel-precise technical drawings and intricate illustrations.
Midjourney just launched V8.1, bringing back its iconic aesthetics with native 2K HD rendering that is 3x faster and 3x cheaper than V8. Its full-quality 1K mode is now faster than V7's draft mode, alongside revived image prompts, an overhauled Describe tool, and new moodboards and style reference features.
Microsoft just rolled out MAI-Image-2-Efficient, a high-throughput text-to-image model built for production scale. It runs 22% faster than its predecessor while slashing pricing by nearly 41% to $19.50 per million image output tokens. Developers can access it now in Microsoft Foundry and MAI Playground, with integrations headed to Copilot, Bing, and PowerPoint.
ByteDance and CUHK researchers just introduced OmniShow, an end-to-end video model built specifically for human-object interactions. It accepts product images, character photos, audio tracks, and pose skeletons all at once, generating up to 10-second marketing clips with synced lip-sync and hand movements. The code is currently under internal review on GitHub ahead of an open-source release.
Researchers just introduced LPM 1.0, an AI model that turns a single photo and audio input into an interactive, real-time visual avatar. Rather than simple lip-syncing, it captures natural idle behaviors, gaze shifts, and pauses while keeping the character's identity stable for up to 45 minutes. It offers a glimpse into true visual conversational partners, though it remains a closed research project for now.
Alibaba just introduced Prompt Relay, a training-free technique that lets video models jump between radically different scenes without messy visual bleeding. By routing time-stamped prompts directly inside cross-attention layers, it keeps each action strictly in its designated slot while smoothing the handoff. The implementation details are up, with full code rolling out shortly.
HeyGen just dropped Avatar V, an AI tool that captures your likeness in just 15 seconds and locks in character consistency across every generated clip. You can change outfits, styles, and settings while keeping your core identity intact. It tackles one of AI video's toughest pain points: preventing faces and characters from morphing between scenes.
CapCut just rolled out its Dreamina Seedance 2.0 AI video generation model to U.S. users across mobile, desktop, and web platforms. Every user gets one free trial to test the new capabilities, alongside a limited-time 90% discount on the first month of CapCut Pro. The rollout is progressive, so access may take a brief period to appear for everyone.
A mysterious new video model called HappyHorse-1.0 just debuted at #1 across both text-to-video and image-to-video leaderboards. Early testing highlights its ability to generate seamless multi-shot clips and accurately follow detailed directional prompts. Why it matters: Dual dominance on major benchmarks indicates a serious new contender has quietly entered the heated AI video race.
Runway just brought Seedance 2.0 to its platform, letting creators generate full multi-shot video sequences complete with sound design and dialogue. It takes text, images, video, or audio as prompts. The catch: it is currently limited to Unlimited and Enterprise users outside the US.
Researchers just introduced Vanast, a unified framework that merges virtual try-on with human motion. You feed it a single portrait, clothing images, and a pose video, and it generates a fluid animation of the person wearing the new outfit while keeping their face and body shape intact. Code is expected to drop soon.
OpenAI's next-gen image model, GPT-Image-2, has reportedly leaked onto the Arena platform under several tape-themed aliases. Developer Pieter Levels noted the mystery models demonstrate exceptionally strong world knowledge and text rendering, potentially beating Nano Banana Pro. OpenAI has yet to officially confirm the testing.
Researchers just introduced MMPhysVideo, a new framework designed to stop AI-generated videos from warping and defying physical common sense. Instead of guessing the next frame purely by pixels, it combines geometry and motion trajectories into unified multimodal training without bloating inference latency, and the team plans to fully open-source the project.
Google just added Veo 3.1 video generation to Google Vids, giving all personal accounts 10 free video generations each month. Paid Google AI Pro and Ultra subscribers also get custom music generation via Lyria 3 and directable AI avatars, along with a Chrome screen recorder and direct-to-YouTube publishing. It turns Google's business-focused video suite into an accessible creative hub for casual users and power creators alike.
Alibaba just launched Wan2.7-Image, a unified model that combines image generation, editing, and understanding. It renders print-quality text across 12 languages, offers precise facial and HEX color controls, and outputs up to 12 consistent images per prompt, solving a major headache for multi-panel creators.
Alibaba just rolled out Wan 2.7 Image, a unified model that handles both generation and precision editing. It focuses on lifelike human faces with natural variation, lets creators lock brand palettes using up to eight hex codes, and renders multi-language text up to 3,000 tokens. Why it matters: Pumping out up to 12 consistent shots in a single prompt makes storyboards and marketing assets far easier to build.
Google just launched Veo 3.1 Lite, its most cost-effective video generation model yet. It delivers the same speed as Veo 3.1 Fast at less than half the cost, supporting text-to-video and image-to-video at up to 1080p. Developers can access it now via Google AI Studio and the Gemini API, with price cuts for Veo 3.1 Fast slated for April 7.
LongCat just launched LongCat Audio Edit, a diffusion-based TTS model capable of cloning anyone's voice with just a few seconds of audio. It comes in a 3.5B parameter flagship version and a 6GB lightweight variant that fits comfortably on consumer GPUs.
ByteDance just introduced DreamLite, a compact 0.39B diffusion model that handles both text-to-image generation and image editing directly on smartphones. It can generate a 1024×1024 image in around 3 seconds on an iPhone, swap styles, and remove objects completely offline. While fine details still lag behind massive server models, it brings fast, private image manipulation right to local devices.
Researchers just introduced TokenDial, a framework that lets users tweak text-to-video attributes like smoke volume or dance motion using a continuous slider. By learning token offsets without retraining base models, it replaces prompt guesswork with predictable, fine-grained control over both visuals and dynamics.
