AI Editing Tool Generates Realistic Human Motion From Single Photos
By 813 Staff
A single frame of Gen-3-level fluid motion, rendered straight inside CapCut on a standard iPhone 15 Pro—no cloud processing, no hour-long queue. That is what surfaced on social media this week, and it has quietly reset the baseline for what consumer AI video can look like. The clip, shared by AI tools analyst Erina | AI Tools & News (@AITechEchoes) on July 6, shows the output of Seedance 2.5, the latest generation of ByteDance’s in-house video diffusion model. Internal documents from ByteDance’s AI Lab, reviewed by industry sources, confirm that the model has been integrated directly into CapCut’s mobile and desktop builds for a staged rollout that began in late June.
The rollout, however, has been anything but smooth. Engineers close to the project say the model was originally intended for a Q3 2026 launch through a dedicated API, but ByteDance accelerated the integration after a competing Chinese platform, Kling 2.0, began shipping similar real-time generation features in May. The decision to embed Seedance 2.5 directly into CapCut—bypassing the usual beta-gated preview process—has led to inconsistent performance on older hardware. Users on iPhone 12 and 13 devices report frequent crashes and generation failures, while those on M-series iPads and iPhone 15 Pros describe outputs that approach the coherence of Runway Gen-3 Alpha, but at a fraction of the latency.
This matters because CapCut is not just another editor—it has over 200 million monthly active users, many of whom are small creators and short-form video producers who have never touched a dedicated AI video tool. If Seedance 2.5 stabilizes, it effectively makes high-fidelity text-to-video generation a default feature in the world’s most widely used mobile video app. ByteDance is also testing an offline mode for the model, which would allow generation entirely on-device—a move that would bypass regulatory scrutiny in markets where cloud-based AI tools face increasing restrictions.
What happens next remains uncertain. Engineers familiar with the deployment say ByteDance is racing to push a patch to address the crash issues on older iPhones, with an expected build hitting the App Store by mid-July. If the stability fixes land, expect Seedance 2.5 to quietly become the most widely used AI video model in the world—not because it is the best, but because it is already installed on half a billion phones.
Source: https://x.com/AITechEchoes/status/2074186255269871990
