1. Hard cuts, not dissolves
When two similar shots are joined with a crossfade, the middle frames show both at once: two puppies on top of each other, a face floating over a tap. The owner had asked for the opening to be “smooth”, and the maker answered with dissolves, which made it worse. Cut on a finished action instead: end the first shot after the drop lands, start the next on the look.
2. An ambience bed under the whole episode
Generated clips often come with very quiet audio, and a still or silent shot can come with none at all. Joined end to end, that produces moments of total digital silence, which a viewer hears as a fault. Lay one continuous bed of room tone or outdoor ambience under the entire episode, about 30 dB under the dialogue, and put speech and effects on top.
3. Lock each clip’s audio to its own picture
One pilot drifted out of lip sync by up to 2.7 seconds by the end. The cause was assembly: the audio was joined as a separate stream, and tiny length differences in each piece added up. The total length still matched, so a simple length check did not notice.
The fix is a habit: trim or pad every clip’s audio to exactly its own video length before joining, and join the clips with picture and sound together. Lesson 6 shows the check that catches this automatically when the maker also delivers each clip on its own.
4. Loudness and delivery
- Mix to about −16 LUFS integrated, with true peaks at or below −1 dBTP. Streaming platforms turn loud uploads down but rarely turn quiet ones up.
- Keep a version without music. It makes the speech check more reliable and lets music change later.
- Titles sit fully inside the frame, with a margin.
- Deliver each episode with a change log (shot, what changed, time in the new version), an edit list and the per-shot clips.