Every one of these has a cost attached — in credits, in hours, or in views that never came. Ordered by how much money they typically burn.
1. Budgeting at list price
2. Rendering drafts on a flagship model
3. Comparing credit plans without converting to dollars
4. Paying for audio you'll never use
5. Stacking two camera movements
6. Writing a scene instead of a shot
7. Skipping the lighting phrase
8. Using one prompt style for every model
9. Prompts past ~90 words
10. Building long videos clip by clip
11. Publishing similar-looking videos back to back
12. Mastering generation, skipping distribution
Nine of these come from treating AI video like traditional filming — where you set up a shot once and it works. Generation is probabilistic: you're buying lottery tickets, and the prompt decides your odds. Every fix above is really the same instruction — reduce what the model has to guess.
Disclosure: some links may be affiliate links, at no extra cost to you.
Budgeting at list price. Because roughly three generations are needed per usable clip, projects planned at the advertised per-second rate come in at about triple the estimate.
Most often a missing lighting phrase. Naming one specific lighting condition — golden hour, neon on wet streets, pale moonlight — does more for perceived quality than any other single word in the prompt.
If it looks like the previous one and was published soon after, the algorithm may treat it as repetition and suppress it. Vary the visual format and leave 2–3 days between similar uploads.