12 product clips with 12 voice lines: one join, one render, 15 cents

A 60-second reel of twelve product clips, each with its own spoken line, costs $0.15275 in Sume jobs once the clips exist: $0.04275 for the voice, $0.01 to join twelve takes into one audio track and $0.10 for a one-minute render. The figures assume twelve lines of 75 characters, 900 in total. Per-job fees are rounded in cents at the catalog level, so treat the fifth decimal as arithmetic and not as an invoice.

Why one voice job per line

Each line gets its own TTS job so a bad take can be redone alone. The jobs share one preset: same voice selector, same generation_config, same language, same output_format. Run them in parallel, each with its own Idempotency-Key, and collect the audio URLs. Timeline audio concat joins up to 20 parts in a single job, so twelve fit with room for a music bed or a closing line.

The ledger

Prices are from the public Sume catalog; video generation for the clips is not included.

Fitting each line to its clip

The reel only works if every spoken line is no longer than its clip. Request the audio, read its duration, and compare it with the clip length. If a line runs over, shorten the words first; changing speed is a second choice, since generation_config.speed ranges from 0.6 to 1.5 and a fast read tires listeners. A line of 75 characters at a typical 15 characters a second runs about 5 seconds, which is also a reasonable clip length for twelve shots in a minute.

Timeline then lets you place the clips and the joined audio on one render. Keep the total at or under 60 seconds: 60.0 seconds is a single billed minute, while 60.1 seconds is two, $0.20, which would raise the whole ledger to $0.25275.

What would change the number

Longer lines raise only the TTS part, at 4.75 cents per thousand characters. Every redone take adds its own characters and one more $0.01 join. A music bed adds $0.125 per generation. Burned-in captions add the caption job on top. The fixed pieces, the join and the render, stay the same until you cross 20 parts or one minute.

Checks before you render

Read every returned audio duration and add them. If the sum is 58.4 seconds the render is one minute; if it is 61 seconds you pay for two. Keep a small margin of two or three seconds for the gaps between lines. Check that every clip URL and audio URL belongs to your workspace, since Timeline works only with Sume-hosted media, and that all twelve lines used the same preset so the voice does not drift between shots.