Here’s what changed this week in AI tools: pricing shifts, new features, and platform updates worth knowing about if you use AI in your marketing or content workflow.
Canva: Craft Image Model Arrives in Dream Lab and the API
Canva released Craft, its own image model, which plans a composition before rendering and outputs up to 2K. Magic Layers can also turn images into editable elements. If you build campaign visuals in Canva, you may need fewer fixes, but still check text, offers and claims before publishing.
Source: Lilach Bullock
ElevenLabs: Realtime Translation and Video to Music
ElevenLabs added a Realtime Translation API for enterprise workspaces and a Video to Music API that generates a soundtrack and matching sound effects for uploaded video. It also shipped WebVTT subtitles for dubbed languages. This matters if you localize video content or want quick background music without licensing headaches.
Source: Releasebot
Runway: More Image and Video Models, Plus Draft Mode
Runway added Seedance 2.5 Draft Mode for fast 480p drafts on all plans, and its API picked up Kling video models and Ideogram 4.5 for images with legible text. If you test lots of video ideas, drafts let you iterate cheaply before spending credits on a final render.
Source: Releasebot
VideoGen: Model 1 Targets 10-Minute Multi-Shot Videos
VideoGen unveiled its first video model, which creates multi-shot videos up to ten minutes long with consistent characters and locations. Rollout is limited to private partners for now, and final pricing is not announced, though the company claims it will cost over 50% less than comparable models. It is also opening applications for a $1 million pool of free credits.
Source: GlobeNewswire
Midjourney: Thinking Mode and Shared Folders in Alpha
Midjourney is testing a Thinking Mode that reruns an image with added reasoning to improve prompt accuracy and typography. Its alpha site also added shared folders with Collaborator and Viewer roles. If your team builds visuals together, shared folders make review easier, but sharing does not change image privacy.
Source: Releasebot
Mistral: Voxtral 3 Speech Models for Transcription
Mistral announced Voxtral 3 speech models and a voice-agent SDK, with vendor-stated support for 50+ languages, speaker identification and vocabulary hints. If you transcribe calls or video content, it is worth testing against your current tool on your own brand and customer names. These are vendor claims, not independent accuracy results.
Source: Lilach Bullock