I made 93 vertical Veo clips under a $75 cap, with zero failed jobs. The interesting part was not the model call. It was refusing to let the batch become a mess.
Every clip went into a manifest before the first paid request. The manifest held the prompt, seed, filename, scene role, model, duration, resolution, and maximum cost. Each one ran four seconds at 1080×1920, with audio and on-screen text disabled.
At the active rate, the most the batch could cost was $74.40. The runner would not cross the approved $75 cap.
It only counted missing files. If a run stopped, it resumed the batch instead of paying to recreate valid clips. It waited through quota pressure and kept the selected model fixed.
The prompts asked for one person, one room, one action, and one continuous shot. Clips started and ended during movement so they would cut together. They also ruled out text, logos, frozen poses, changing furniture, extra people, damaged hands, and sudden camera jumps.
The 93 clips became 12 short edits for three products. A local FFmpeg script made those edits with uneven shot lengths and hard cuts, with no extra generation cost.