Pet video workflows.
Podcast & interview.

Goal and workflow

The podcast aimed to turn two pets into conversational characters, using uploaded photos and a topic to generate a short podcast-style video. The workflow needed to bring together dialogue, character visuals, voiceover, and editing.

The workflow starts with pet photos and a topic, develops dialogue and storyboard imagery, prepares TTS audio and video clips, and combines video, sound, and subtitles.

Voice selection and the time each line occupies affect the viewing experience. These were the two problems I focused on.

Listen first, then constrain selection

Voice selection was one of the hardest parts. Letting the workflow find a voice it considered suitable made it difficult to express my own listening preferences.

I listened to voices individually, compiled a shortlist, and required the workflow to choose only from those voices.

This separated selection into two steps: I handled listening and curation, then the workflow chose within that range. Human judgment became an explicit constraint.

Separate generation from edit length

The voiceover and the pets' mouth movements did not line up in time.

The video tool I used required clips of at least five seconds, while dialogue lines varied in length. I controlled dialogue length through the script, generated five-second clips, and trimmed them during editing.

This let the retained clip length follow the voiceover. The approach addressed duration and pacing; it did not in itself establish precise, phoneme-level lip synchronization.

Goal and workflow

The pet interview is a separate workflow I independently built from a reference template. It takes two pet photos and an optional description to create an interview with character-specific dialogue and voices.

Photos and descriptions → character features → interview script → voice and shot-duration checks → video and subtitle assembly.

Extract character features

I instructed the model to extract animal features systematically and produce standardized visual descriptions as references across shots.

Compare duration and revise dialogue

The interview workflow compares TTS duration with shot duration, then shortens dialogue and checks again. This is distinct from generating and trimming five-second clips in the podcast; duration matching does not establish precise lip synchronization.