DEV Community: Marcus Vale

Why AI video prompts fail without a brief

Marcus Vale — Wed, 09 Sep 2026 03:43:55 +0000

I used to think bad AI video was a prompting skill issue. Better adjectives. More camera terms. Negative prompts stacked like a grocery list. Some of that helps. Most of my failures weren’t the prompt though. The prompt had nothing solid to aim at.

You’ll recognize this. Client says “make a premium product video.” You write a long prompt. Model spits out something shiny. Everyone squints. Nobody can say what’s wrong, so you tweak the prompt again. An hour later you have eight almost-clips and zero agreement.

That’s not a model problem. That’s a missing brief.

A prompt describes how to render a moment. A brief decides what the video owes the viewer. If you skip the brief, every regeneration is you guessing the job in prettier language.

What actually fails

Revisions attack the wrong layer
People change “cinematic lighting” when the real issue is “this doesn’t prove setup is under 10 minutes.” Lighting wasn’t the miss. The brief never named the miss.
Storyboard gets skipped because prompting feels faster
It isn’t. A five-beat scribble would’ve shown the hole before you spent credits.

The order that stopped me wasting credits

Brief first. Storyboard second. Prompt last.

Brief = audience, one job, proof on screen, do-nots, length/format, revision rule.
Storyboard = ugly beats that hit that proof.
Prompt = how each beat looks.

I still write prompts. I just refuse to write them into a vacuum.

Tiny failed example

Prompt-only: “Cinematic demo of an AI editing app, sleek UI, shallow depth of field, premium tech ad, 40 seconds.”

What we got: beautiful fake screens. Client: “Where’s the actual edit?”

Open a note and answer these before you touch the model:

Who is this for, in a real moment?
What single job does the video have?
What would I see that makes me believe it?
What are we not doing?
How long, what format, how many revision rounds?

If you can’t answer those, you’re not bad at prompting. You’re early.

Related (AI video cluster only):

Brief template hub: https://medium.com/@casey.vale/ai-video-brief-template-what-i-fill-in-before-i-generate-8d81244ff992- Paper before generate: https://dev.to/marcusvale/i-stopped-generating-ai-video-until-the-storyboard-was-ugly-on-paper-fc9
How to brief a client: https://dev.to/marcusvale/how-to-brief-a-client-for-ai-video-1h59

I helped build a short AI video brief + storyboard kit if you want the template in one place: https://scienceartist.gumroad.com/l/ai_video_creator_brief_storyboard_kit
Brief-first version of the same ask:

Audience: freelance editor between client rounds
One job: believe a rough cut becomes usable in one pass
Proof: timeline scrub, messy → clean
Do not: UI fireworks, robot voice Then the prompt for the proof beat was boring and specific. It worked on take two.

If your prompts keep “almost” working

The prompt is doing product-manager work
“Show it’s fast, powerful, and easy” is three jobs. A prompt can’t prioritize. The model will average them into vague glow.
There’s no proof beat
Prompts love aesthetic. Clients approve proof. If you never wrote “viewer should see X on screen,” you’ll keep getting mood and missing the receipt.

Free client brief worksheet: https://scienceartist.gumroad.com/l/free-client-brief-worksheet — if you want the full AI video brief + storyboard kit after: https://scienceartist.gumroad.com/l/ai_video_creator_brief_storyboard_kit

Freelancer intake form before deposit

Marcus Vale — Wed, 09 Sep 2026 03:01:37 +0000

I used to take deposits on a vibe call. Client sounded clear. Money hit. Then the real project showed up in email #6 and it wasn’t what I priced.

Now the order is fixed: intake form → short scope confirm → deposit. No deposit before intake. Sounds strict. It’s cheaper than rework.

Why before deposit matters

Deposit locks commitment on both sides. If intake isn’t done, you’re locking fog. People also get nicer about filling a form when they know payment is next, not three weeks into Slack chaos.

The form (short on purpose)

Intake (before deposit)

Project in one sentence:
Success looks like:
Must include / must not:
Deadline + hard dates:
Who approves:
Revision rounds in this deposit:
Links / access I’ll need on day one:

If they can’t answer those, they’re not ready to pay. That’s useful information.

What I say when they want to “just pay and start”

“Happy to. Intake takes 5 minutes, then I send the deposit link.” Most fill it. The ones who push hard were usually going to be scope problems anyway.

One paid lesson

Took a deposit for “landing page copy.” Intake would’ve caught “also need email sequence + two PDFs.” I ate it to keep the relationship. Relationship didn’t renew. Intake would have either expanded the quote or killed the job early.

Deposit isn’t the start of work. Intake is. Deposit is the gate after clarity.

I helped build a freelancer client ops bundle if you want intake + briefs + delivery in one stack: https://scienceartist.gumroad.com/l/freelancer-client-ops-bundle

Wedding content questionnaire for creators

Marcus Vale — Wed, 09 Sep 2026 02:58:05 +0000

Most wedding gigs go sideways in the same place. Couple thinks you “just know” the day. You think they’ll send details. Nobody sends a real questionnaire, so you find out about uncle speeches and sparkler exits an hour before they happen.

I started sending a short questionnaire before I lock the rate. If they won’t fill it, they’re not ready to book.

What I ask (copy if you want)

Wedding content questionnaire

Couple names:
Wedding date + city:
Ceremony start time / reception start:
Must-get moments (list 5 max):
Moments we can skip:
Family / VIP names we should know:
Any ex / conflict people to avoid in frame:
Shot list constraints (religion, privacy, unplugged?):
Delivery: highlight length, vertical vs horizontal, deadline:
Who approves the final cut:

That’s enough. Long forms get abandoned. This one gets answered.

Why creators need this more than photographers

Photo CRMs already have intake. A lot of wedding content creators are still running on Instagram DMs. That’s how timelines slip and “quick reels” become unpaid second shoots.

The questionnaire is also your scope lock. If “must-get” grows after deposit, that’s a change order, not a favor.

One real miss

I once skipped the VIP names field. Shot a beautiful toast. Wrong uncle. Couple was polite and also never referred me. Now that field stays.

How I send it

Notion or Google Form. Same questions. Due before the planning call. I don’t run the call blind anymore.

If your wedding work feels chaotic, it’s usually missing intake, not missing gear.

I helped build a wedding creator workflow kit (questionnaire, timeline, delivery pack) if you want it in one place: https://scienceartist.gumroad.com/l/wedding-content-creator-workflow-kit

Brand outreach scripts that don’t get ignored

Marcus Vale — Wed, 09 Sep 2026 02:55:34 +0000

Most brand DMs die in the same spot. “Hey love your content, would love to collab.” Creator feels nothing. Brand follows up once, then ghosts the idea.

I did this for months. Open rate fine. Reply rate garbage.

What changed it wasn’t being funnier. It was leading with an angle the brand can say yes to without a meeting.

Why cold scripts get ignored

You’re asking them to invent the campaign. Their inbox is already full of creators who “make great UGC.” Same pitch, different face.

If your first message doesn’t contain a usable idea, you’re unpaid strategy work. They skip.

Script shape that gets replies (for me)

One line on who you make content for (specific, not “Gen Z”)
One angle tied to their product or a recent post
One proof you’re not random (a result, a niche, a format you own)
One soft ask (15-min fit check / want the 3 angles in a doc)

Keep it under ~90 words. Longer reads like a proposal they didn’t request.

Example that actually got a yes

Bad:
“Hi! I’m a UGC creator and I’d love to work with you. I can do TikTok + Reels. Rates flexible!”

Better:
“Hey — I make short demos for indie SaaS (mostly ‘setup in under 10 min’ proof clips). Saw your new onboarding flow. Angle I’d shoot: screen + face, timer on, first publish in one take, no feature dump. Want me to send 2 alternate hooks in a doc, or not a fit?”

They replied same day. Not because it was clever. Because they could picture the video.

Do-nots for the outreach itself

Don’t attach a media kit in message one
Don’t list every platform you use
Don’t say “passionate about your brand” if you just found them
Don’t send the same script to 40 brands with the name swapped (they can tell)

Follow-up

One bump after 4–5 days. Same angle, shorter. If silence, move on. Chasing kills the frame that you’re selective.

I keep a dumb tracker: brand, angle sent, date, reply, next step. Without it I either spam or forget the ones that said “next quarter.”

I helped build a UGC pitch kit (angles, DM/email scripts, follow-up tracker) if you want that stack in one place: https://scienceartist.gumroad.com/l/ugc_brand_pitch_tracker_scripts

UGC brief do-not list examples (the ones that actually help)

Marcus Vale — Wed, 09 Sep 2026 02:47:46 +0000

“Don’t be salesy” is not a do-not. Creators can’t argue with it, so they play safe and the video dies.

A useful do-not list is short and specific. Max three. Stuff you could reject in review without a taste fight.

What I put on real briefs

Bad:

Don’t sound salesy
Keep it authentic
No cringe
Be natural

Those are vibes. They create fear, not clarity.

Better (steal these if they fit):

No logo end card
No “link in bio” spoken out loud
No medical / before-after skin claims
No competitor names
No stock B-roll of smiling strangers
No three CTAs in one video
No script that starts with the product name in the first 2 seconds
No green screen
No fake “I’ve been using this for months” if the creator just got the product

Pick three. Not nine.

Why three

I watched a brand dump twelve bans into a doc. Creator froze, shot the blandest talking head possible, brand said it lacked energy. The do-not list ate the brief.

Three hard nos. Everything else is a preference in a different section, or it doesn’t belong.

How it sits next to the rest of the brief

Do-nots only work if the job and proof beat are clear. “No logo end card” is useless if nobody wrote what the video is supposed to prove. Fill audience, one job, proof on screen, then the three bans, then length/format.

Quick example

Brief do-nots for a UGC skincare clip we actually shot:

No dermatologist voice
No before/after medical language
No end card

Proof was texture on a bathroom counter in daylight. That combo got approved on round one. The earlier version had “don’t be salesy” and went three rounds.

If your do-not list reads like a mood board, rewrite it until each line could be a binary fail in review.

I helped build a UGC pitch kit (angles, scripts, follow-up tracker) if you want that lane in one place: https://scienceartist.gumroad.com/l/ugc_brand_pitch_tracker_scripts

How to brief a client for AI video

Marcus Vale — Wed, 09 Sep 2026 02:44:09 +0000

Most client AI video briefs I get are either a vibe paragraph or a 10-page doc nobody reads. Both end the same way. You generate something pretty. They say it’s not what they meant.

Here’s the short version I use now. Send the client a one-page brief. Fill it with them on a call if you have to. Don’t generate until it’s done.

Copy-paste (client-facing):

AI video brief (client)

Working title:
Who is this for (real situation):
One job for the video (one sentence):
What should we see on screen that proves it:
Must include:
Must not include (max 3):
Length + format (e.g. 9:16, 40–55s):
References (optional, 1–2 links max):
Revision rule: ___ rounds, notes due within ___ hours, only against this brief
Sign-off: name / date

That’s the brief. Not the storyboard.

People mix those up a lot. Brief = what success is. Storyboard = the beats you’ll shoot or generate to hit it. If the client starts rewriting shot 3 before they’ve agreed the one job, pause. Lock the brief first. Storyboard second. Generate third.

Revision rules belong in the brief or you’ll eat infinite “one more take” energy. I write it blunt: one revision round, notes only against the signed lines above, no new must-includes after sign-off. Sounds harsh. Saves both sides.

Example. A client wanted “an AI tool video that feels premium.” We filled this instead. Audience: editors between freelance jobs. One job: believe a rough cut cleans up in one pass. Proof on screen: timeline scrub, messy to usable. Must not: glowing UI porn, robot voice jokes. One revision. We storyboarded five beats after that. First assembly mostly landed. The old way would’ve been eight regenerations and a vague “make it punchier.”

If they won’t fill the brief, they’re not ready to generate. Do the boring page first.

AI video brief template (Medium hub): https://medium.com/@casey.vale/ai-video-brief-template-what-i-fill-in-before-i-generate-8d81244ff992
Paper before generate: https://dev.to/marcusvale/i-stopped-generating-ai-video-until-the-storyboard-was-ugly-on-paper-fc9

I stopped generating AI video until the storyboard was ugly on paper

Marcus Vale — Tue, 08 Sep 2026 19:19:05 +0000

Last month I burned through a chunk of credits on a “simple” product clip. Client said make it feel real. I opened the generator first, because that’s the fun part. Got eight almost-good takes. None of them survived review. The note was basically: cool, but what’s the point.

That one stung because the footage wasn’t bad. I just never locked what the clip had to prove.

So I flipped the order. Paper first. Model last.

Here’s what I actually do now, and it’s messier than a framework slide.

I write the brief like I’m texting a tired editor. Who’s this for, in a real moment. What is the one job. What would make me believe it if I was scrolling half-asleep. Three things we will not do. Length and ratio so nobody “surprises” me in the cut.

Then I storyboard badly on purpose. Sticky notes, or a dumb table in Notion. Hook. Mess. The fix happening. The proof you can see. Done. If I can’t draw the proof beat without jargon, I don’t generate yet. That rule alone saved me from another vanity render last Tuesday (nice lighting, zero point, client ghosted the cut for two days).

Only after that do I touch the model. One beat at a time. If proof fails, I regen that beat. I don’t throw the whole timeline in the trash like I used to.

Failed shots used to feel random. They’re usually me skipping the ugly paper step.

If you’re about to open the generator with a vibe and a prayer, close it. Scribble the proof beat first. Your future self (and your credit balance) will be less annoyed.