Video Prompt Mistakes: 8 That Waste a Render
The most common video prompt mistakes are a vague ask, no length, ten requests in one message, and rendering before looking at the plan. Each has a short fix: say the story, give the seconds, send one or two notes at a time, and ask for the timeline or the preview sheet first. None of them needs a longer prompt, only a more exact one.
These are written for editing with Claude or ChatGPT through Cut Express, where your AI writes the edit and Cut Express renders the MP4. Each mistake below has a prompt that fixes it. Fill the blanks in capitals.
Which mistakes cost the most?
| Mistake | What happens | The fix |
|---|---|---|
| A vague ask | A generic video | Say the story in one sentence |
| No length | Your AI guesses | Give seconds |
| Ten requests in one | Some get lost | One or two notes per version |
| Skipping the preview | You pay for a render to find a typo | Ask for the timeline or the sheet |
| “Redo it” | A different video, not a better one | Say the scene and the change |
| Asking for facts it lacks | Blanks, or a question back | Give prices, names and numbers |
| A voice before the script is final | A voice you have to make again | Approve the words first |
| No style on a drawn video | An edit, or the wrong look | Name the style in the first line |
Mistake 1: is “make it good” a prompt?
It is a wish. Compare the two.
Vague:
Make a good video from my clips.
Specific:
Make a 30-second 9:16 reel from my clips about opening day at my bakery:
the queue outside, the first tray, a customer's reaction, the sold-out
sign last. Fast cuts, warm look, music from the track I uploaded.
And the fill-in version:
Make a [LENGTH IN SECONDS]-second [FORMAT] video from my clips about
[WHAT HAPPENS, IN ONE SENTENCE]. In this order: [FIRST], [THEN], [LAST].
[PACE], [LOOK] look, [MUSIC].
Why it works: an order of events is something an editor can cut to. The full anatomy is in how to write a video prompt.
Mistake 2: what happens with no length?
Your AI picks one, and it is rarely yours. Length also decides the pace: the same footage is a different video in 15 seconds than in 60.
Cut this to [LENGTH IN SECONDS] seconds. If the footage does not fill it
well, tell me and suggest a shorter length instead of padding it.
Why it works: it gives the number and says what to do when the number is wrong. For short-form, see how long a reel should be.
Mistake 3: why not ask for ten changes at once?
Because some get lost, and when the result is worse you cannot tell which change did it. Send the two that matter most.
v2, two changes only: [CHANGE 1] and [CHANGE 2]. Keep everything else
exactly as it is.
Why it works: “keep everything else” stops a note from turning into a new edit.
Mistake 4: why look before the render?
A list of scenes takes seconds to read. A finished video with a wrong line needs another render to fix.
Before you render, show me the timeline: each scene with its time, every
text on screen, the captions style and the sound. Wait for my OK.
Why it works: it turns the render into the second step, not the first. For long footage, ask for the moments first:
Inspect [LONG VIDEO] with moments and a transcript. List the
[NUMBER OF CLIPS] best moments with their times and first line. Do not
render anything until I pick.
Why it works: you choose from real moments, as in shorts from a long video. For a drawn video, the same idea is “show me the preview sheet before you render”, used in every one of the drawn style prompts.
Mistake 5: how do you give notes on a version?
Not with “redo it, it’s not good”. Say where and what: the scene or the second, then the change.
v[NUMBER] notes: cut scene [SCENE NUMBER] to [NEW LENGTH] seconds. Move
the captions [WHERE] so they do not cover [WHAT THEY COVER]. Bring the
music down under the voice. Swap the opening for [THE NEW OPENING].
Keep everything else.
Why it works: each note has a place and a change, like notes to a human editor. Every version is kept, so a note that does not work costs you nothing but the try.
Good notes sound like this:
- “Scene 3 is too long, cut it to 2 seconds.”
- “The caption covers my chin, move it down.”
- “Start on the laugh instead of the hello.”
- “Same edit, but music half as loud.”
Mistake 6: can the AI fill in your facts?
No, and it should not. Prices, results, names and quotes come from you. Give them in quotes so they appear as typed.
Use these facts exactly and add none: product "[YOUR PRODUCT]", offer
"[OFFER]", price "[PRICE]", result "[THE RESULT]". If you need a fact I did
not give, ask me.
Why it works: it draws a line between what you supply and what your AI decides. This matters most in ads: see prompts for product ads.
Mistake 7: when should the voice be made?
After the script is final. A voice-over is made from the exact words, so a changed sentence means making it again.
Write the script for [TOPIC], about [NUMBER OF WORDS] words, and show it
to me. Do not make the voice until I approve the words. Then make it:
[VOICE].
Why it works: the approval comes before the step that spends credits. More in AI voiceover for video.
Mistake 8: what if you forget to name the style?
For a drawn video, the style is what tells your AI to draw instead of edit. “A retro video about toasters” could go four ways. “A pixel art video of how a toaster works” goes one way. See prompts for pixel art videos and the styles page.
FAQ
Is a long prompt a mistake?
No. Long and exact is fine. Long and vague is the problem.
How many notes per version?
One or two, three at most.
What if version 2 is worse?
Go back: “use v1 again and only change the captions”. Earlier versions are kept.
Where can I see prompts that avoid all of this?
The prompt library, the prompt guide and the pillar post on video editing prompts. For each AI, see ChatGPT video editing prompts and Claude video prompts.
Fix the prompt, not the render
Cut Express is the video editor you drive from Claude or ChatGPT: describe the video, check the plan, give notes, get the MP4. See the plans.