InfiniteTalk

Based on the available evidence, InfiniteTalk looks worth tracking, but the adoption call should be “best for people comfortable with open-source workflows,” not “default polished SaaS.” The heat proof is strong: many X reposts, summary posts, and “Meituan open-sourced it” mentions drove wide attention, and references to ComfyUI plus WaveSpeedAI suggest meaningful interest in the AI video community. Usability proof is narrower. It comes mainly from two Zhihu hands-on articles, a few tutorial-style shares, and platform demos, which support that it can produce concrete outputs such as single-speaker avatars, two-character conversations, and audio-driven talking videos, but public samples are still relatively demo-heavy.

Its real job is not text-to-movie or a general AI video generator. A better comparison is a long-form talking-avatar and lip-sync pipeline. Across the evidence, the recurring capability is taking a photo or a person video plus audio and turning it into a speaking avatar video. Some examples further emphasize two-character interaction, multi-turn dialogue, synchronized facial, head, and body motion, and longer-duration output.