Aulafy
Courses/Generative AI: image, voice and video/Local video with Wan and ComfyUI

Local video with Wan and ComfyUI

Generative video is tempting, but memory-intensive and easy to get out of hand. For education, start with short clips, simple shots, and a clear role within the lesson.

  • Design short clips that explain a specific idea.
  • Understand when to use text-to-video or image-to-video.
  • Combine image, voice, and subtitles into a reviewable result.

Scene plan before the model

TerminalCode
{
  "escena": "vector database",
  "duracion_segundos": 4,
  "formato": "16:9",
  "entrada": "imagen aprobada rag-hibrido-portada.png",
  "movimiento": "paneo lento hacia los documentos citados",
  "prohibido": ["texto pequeño ilegible", "caras", "logos"],
  "salida": "clip-rag-hibrido-4s.mp4"
}

Recommended workflow

  • Generate a still image and approve it.
  • Create a 3 to 5 second clip with minimal movement.
  • Check for deformations, broken text, and visual identity shifts.
  • Add narration and subtitles outside the video model.
  • Export a lightweight version for the web.
TerminalCode
ffmpeg -i clip-rag-hibrido-4s.mp4 \
  -i narracion.wav \
  -shortest \
  -c:v libx264 -crf 23 -preset medium \
  -c:a aac \
  salida-capsula.mp4