Adobe pushes AI video
Adobe unveiled MotionStream, an experimental AI video tool aimed at giving users more precise control over generated video and blurring the line between stills and motion workflows. The company is positioning Firefly and other generative tools as native parts of creative work, which could change how photographers package short-form education and before/after demo content. If these tools make richer motion content easier, creators may shift toward mixed-media product pages and tutorials that combine still edits with short motion demonstrations. (nofilmschool.com) (progressiverobot.com)
Adobe is trying to turn AI video from a slot machine into something closer to a camera rig. On April 10, 2026, Adobe Research showed MotionStream, an experimental system that lets a user drag objects, steer camera moves, and change motion while the video is still being generated. (research.adobe.com) Most AI video tools work like ordering food at a drive-through. You type a prompt, wait, and see what comes out, which is why small fixes like “move the dog left” or “tilt the camera down” often mean generating the whole clip again. (research.adobe.com) MotionStream is built around a different idea: stream the video continuously so the person can intervene mid-shot. Adobe says users can control movement with a cursor and sliders, and the model updates motion in real time instead of treating the clip like a finished export. (research.adobe.com) The research paper behind it says the system is “two orders of magnitude faster” than earlier approaches and can produce “infinite-length streaming” video. In plain English, Adobe is claiming speed high enough that the human hand can stay in the loop instead of waiting for each batch render to finish. (arxiv.org) That control problem has been the weak spot in generative video for more than a year. Adobe’s own Firefly Video Model launch in October 2024 focused on “commercially safe” generation, and by February 2025 Adobe was selling tighter controls like camera angle, start and end frames, and style references. (news.adobe.com) (blog.adobe.com) By April 2025, Adobe had folded image, video, audio, and vector generation into one Firefly app and said more than 22 billion assets had been generated with Firefly in under two years. That tells you where the company is headed: fewer separate “AI tricks,” more one-stop workflows inside the Adobe stack. (blog.adobe.com) Adobe’s latest Firefly product pages now pitch the app as a place to create and edit images, audio, and video, and even use outside models from Google and OpenAI alongside Adobe’s own tools. MotionStream is still a research demo, but it fits that larger plan almost perfectly: make motion another native control, not a separate specialty. (adobe.com 1) (adobe.com 2) That could change a very specific kind of creator work first. A photographer selling presets, a retoucher showing a before-and-after, or a teacher making a 20-second lighting demo can usually make a still image faster than a polished video, so mixed image-and-motion posts have been bottlenecked by time and editing skill. (adobe.com) (nofilmschool.com) Adobe is also pushing harder into mainstream video workflows at the same time. In January 2026 it announced new artificial intelligence tools for Premiere Pro and After Effects, and said 85 percent of Sundance Film Festival premieres were made with Adobe Creative Cloud tools. (blog.adobe.com) (news.adobe.com) So the bet here is not just “people want AI video.” The bet is that if Adobe can make motion as tweakable as a Photoshop layer, the basic unit of creative work stops being a still image or a finished video and becomes something in between. (research.adobe.com) (blog.adobe.com)
