Consolidation Wave: Sora Dies, Veo Democratizes, and the Image Generation Arms Race Heats Up

News

OpenAI Pulls the Plug on Sora: A $1M/Day Money Pit That Couldn't Find Its Audience

Sora was burning roughly $1 million every day in compute costs while active users collapsed from 1 million to under 500,000 — OpenAI killed it to free up GPU capacity for enterprise and coding products. This is the clearest signal yet that AI video as a consumer social product isn't viable at current economics, and it killed a $1B Disney deal in the process.

GPT Image 2 Leaks on Chatbot Arena Under Adhesive Tape Codenames

Three variants of what appears to be OpenAI's next image model — codenamed maskingtape, gaffertape, and packingtape — have surfaced in blind Arena tests, with community testers noting dramatically better text rendering, spatial reasoning, and color accuracy over GPT Image 1. Analyst estimates peg a formal launch at May–June 2026, likely integrated natively into GPT-5 rather than as a standalone model.

Tools

Google Launches Veo 3.1 Lite: Sub-50% the Cost of Veo Fast, Same Speed

Google dropped Veo 3.1 Lite on March 31 — priced at less than half of Veo 3.1 Fast with the same generation speed, available now via the Gemini API and Google AI Studio. With Sora gone, Google is aggressively moving to own the developer video generation stack by competing on cost-per-second, not just quality.

Veo 3.1 Lite Also Ships with Standalone Video Upscaling on Vertex AI

Alongside the Lite model, Google quietly launched a standalone Veo upscaling capability on Vertex AI — letting you enhance any existing video (AI-generated or traditionally shot) up to 1080p and 4K. This is a practical tool for teams who have existing footage they want to bring up to modern quality standards without re-generating from scratch.

Google Vids Gets Text-Directed AI Avatars, Custom Music, and One-Click YouTube Publishing

Google Vids' latest update lets you direct AI avatars with plain text prompts — placing them in scenes and having them interact with products or props — plus adds built-in music generation and direct YouTube publishing. It's evolving quickly from a lightweight editor into a genuine text-to-video production environment aimed squarely at enterprise content teams.

Analysis

Hands-On with MAI-Image-2: Strong Photorealism and Text Rendering, But No Editing Features Yet

Decrypt's hands-on found MAI-Image-2 beats GPT-Image on image quality and text rendering despite sitting lower on the Arena leaderboard — but it's currently text-to-image only with no inpainting, outpainting, or reference image support, and outputs are locked to 1:1 ratio. Worth testing for enterprise brand work, but not yet a Midjourney or Firefly replacement for iterative creative workflows.

Sora's Shutdown Is a Warning to Chinese AI Video Rivals, Not an Opportunity — Bloomberg Analysis

Bloomberg's Catherine Thorbecke argues that Kling, Vidu, and ByteDance's Seedance 2.0 should read Sora's death as a cautionary signal, not an opening — the consumer social video model simply doesn't work at current compute costs. ByteDance has the biggest distribution moat via CapCut's 741M MAUs, but none of these players have cracked the unit economics either.