@laowangbabababa: Shocked! Dr. Qi on Douyin sells a 500k digital human agent per day, and I built it in 2 minutes. Using the Pixelle-Video project, which already has 22k stars. It supports digital human lip-syncing, motion transfer, and image-to-video. Supports ComfyUI, input a topic, from script writing to adding...

Shocked! Dr. Qi on Douyin sells a 500k digital human agent per day, and I built it in 2 minutes. Using the Pixelle-Video project, which already has 22k stars. It supports digital human lip-syncing, motion transfer, and image-to-video. Supports ComfyUI, input a topic, from script writing to adding BGM to final output, fully automated video production. Old Wang deployed it locally and made a short video, which is an illustration-style video. If you need more complex videos, you need to configure cloud models like seedance2, kling, etc. The best part of Pixelle-Video is that it makes video production fully configurable, supporting both locally deployed models and cloud-based large models. Script, visuals, voiceover, and editing are split into four replaceable modules. Behind each module, you can switch models freely, making it very convenient. > Script layer: LLM reads the topic and outputs a structured script with timestamps, each sentence corresponding to a segment of visuals. > Visual layer: Each sentence of the script is converted into an image generation prompt, sent to ComfyUI or directly to DashScope to generate images. Image-to-video and digital human lip-syncing also go through this layer. > Voiceover layer: The original script goes through TTS synthesis, supporting multiple languages and voice cloning, no need to record your own audio. > Compositing layer: Synchronize visuals with the voiceover timeline, overlay BGM, and output MP4. Repo: http://github.com/AIDC-AI/Pixelle-Video… P.S. Think of it and you can produce a video. With a little AI programming, this project can be customized into a digital human agent suitable for various industries.