How to run a one-person studio supplying product videos, UGC ads and review content to brands that pay $1,000 a project or $500 a month — no crew, no office, no stack of subscriptions.
ContentsTap any line to jump
Tens of thousands of people have now learned to make AI video. Almost none of them have been told who pays for it.
So they do the obvious thing. Post it to a faceless account. Hope the algorithm picks it up. Wait for a platform or a sponsor to hand over a fraction of a penny per view, in a queue with a million other people doing exactly the same thing on exactly the same day.
That is not a business. That is a raffle you enter daily.
The other way is quiet, unglamorous, and almost nobody is competing with you for it.
You sell the video to the brand.
Not to a stranger on the internet. To a business that already has a product, already has an ad account, already has money leaving its bank every single month — and is losing that money right now, because it cannot produce creative fast enough to feed the thing it is paying for.
That brand does not want your showreel and does not care which model you used. It wants thirty pieces of usable content this month, thirty more next month, and someone reliable enough that it stops being a thing they think about.
Give them that and they will pay you for it every month until one of you stops trading.
This is a studio, not a lottery ticket. You set a brand up once and produce for it repeatedly — and the second month takes half the work of the first. That is the whole idea. The rest of this book is how.
There are two separate skills here, and almost everybody only ever learns the first one.
Making it. Prompting, lighting, motion, sound, consistency, storytelling. Taught well in plenty of places, and taught here in Part One — nine phases, in order, each one assuming you did the last.
Selling it. Pricing, packaging, who to approach, what to say, what to deliver, how to keep them past month three. Taught almost nowhere, which is why there are tens of thousands of people who can generate a beautiful UGC ad and cannot name a single business that would pay them for one. That is Part Two.
If you already have the craft, skip to Part Two and lose nothing. If you don't, do not skip — every business chapter assumes you can hit the quality bar, and no amount of good pricing survives work that looks fake.
Nine phases, in order. Each one assumes the last. By the end you can hit the quality bar a paying brand expects — which is the entry ticket to everything in Part Two.
Before any prompting, understand the shape of the thing. You have four kinds of model — image, video, voice, language — and they do not compete, they hand off. Language writes the script. Image builds the frame. Video moves the frame. Voice carries it. Almost every bad result comes from asking one model to do another one's job.
Learn the credit economics on day one, because they govern every creative decision you will make later. Cheap models exist for drafting; expensive ones for finals. Using a premium video model to test whether a composition works is how beginners spend a month's budget in an afternoon.
Then build one brand profile — even a fake one, for practice. Brand, product, tone, customer. Everything downstream gets easier once the system knows who it is making things for.
The actual skill is composition — chaining tools so the output of one becomes the input of the next. A brand profile in, a script out. A script in, a voice out. A voice and an image in, a video out. Every bad result comes from asking one model to do another model's job while you watch. The best results come from building the corridor first, then letting the model fill it.
Video quality is mostly image quality with movement added. Get this phase wrong and nothing later saves you.
Write prompts in a fixed order and stop improvising the structure: subject → action → setting → light → lens → mood. Every element you leave out is a decision the model makes for you, and it will make the most average one available.
The specific lever that separates amateur from professional output is light. "Well lit" means nothing. "Overcast window light from camera left, soft shadow falloff" means something. Name the source, the direction and the quality, every time.
Then learn to edit rather than regenerate. Changing one element in an existing image costs a fraction of rolling the dice again, and it is how you keep a look consistent across thirty assets.
Finally, batch. Generate four variations of a composition and pick, rather than generating one and hoping. Selection is a skill and you cannot select from one option.
Default to image-to-video, not text-to-video. Starting from a still you have already approved means you control the composition, the product and the lighting before any motion is introduced. Text-to-video gives the model authority over everything at once, which is why it drifts.
Learn camera language properly, because it is the vocabulary the model actually responds to: slow push-in, orbit, handheld drift, rack focus, static locked-off. Name one movement per shot. Two movements in one prompt produce mush.
The most reliable way to control that movement is with keyframes, not prompts alone. Generate the still first — locked composition, locked product, locked light. Then generate the move from that still. If you need a specific action or landing point, generate an end-frame too and let the model fill the gap between them. Two frames give the model a corridor; one frame gives it a suggestion. The more precise the input, the less the model has to guess, and the less it drifts.
The hardest discipline in this phase is brevity. Five to eight seconds per shot. Longer clips accumulate error — hands wander, faces slide, physics gives up around second nine. Ads are built from short shots anyway, so the constraint costs you nothing and saves you constantly.
Most people obsess over pixels and ship silence. Sound is the single highest-return, lowest-cost thing in this entire book — a mediocre clip with excellent audio reads as real; a beautiful clip with tinny audio reads as fake.
Three layers, always: voice, room, music. Voice carries the message. Room tone — the ambient hum of the space, the fabric rustle, the cup on the counter — is what convinces the ear it is watching a place rather than a render. Music sets the pace.
Pick a voice with intent. Age, accent and energy should match the customer you are talking to, not the one you find pleasant. Preview before you commit, and once you choose a presenter's voice, write the ID down — you are going to need the same one next month.
Then caption everything. Most social video is watched muted, and burnt-in captions are not an accessibility afterthought, they are the primary channel.
This is the phase that decides whether you can charge money. A one-off clip is a party trick; eight clips featuring the same presenter, in the same kitchen, holding the same product, is a deliverable.
Consistency has four locks and you need all of them: face, outfit, place and voice. Miss any one and the viewer feels the join without being able to name it.
The method is reference discipline. Build the person once — face reference, full-body reference, four angles, a couple of expressions — and never generate them from scratch again. Do the same for the location. Save it all against the brand.
The full working system for this, including how it fits into a paid delivery, is Modules 01, 02 and 06 in the technique library later in this book. Learn the principle here; run the process there.
An ad is not a beautiful video. An ad is a beautiful video arranged in an order that sells, and the arrangement matters more than any individual frame.
The structure that works, near enough universally in short form: hook (something happens before the viewer decides to scroll), problem (name the annoyance out loud), proof (show the thing working), offer (what to do and why now).
The first second is not part of the ad — it is the ad. Nothing else gets watched if it fails. Open on motion, on a face mid-sentence, or on the product already doing the interesting thing. Never open on a logo.
Learn the five angles until they are automatic: problem/agitate, before-and-after, social proof, founder-to-camera, offer. Every brand, every niche, forever. You are not looking for a new idea each month; you are looking for this month's version of five ideas that already work.
Once the angles are automatic, the work becomes an inner loop. Generate five rough cuts — one per angle — in the time it used to take to make one. Test them cheaply, read the signal, and put real production budget behind the direction that moves the needle. AI eliminates the cost of the experiment; it does not eliminate the need to judge the result.
The numbers that tell you whether to keep going are simple: thumb-stop rate (first three seconds), playthrough rate (full video), CTR, and add-to-cart. If an angle is not moving at least one of those within the first hundred views, cut it and spend the credits on the next one. Volume of testing is the strategy; the numbers are the filter.
The same ad performs differently on three platforms, and most of that difference is packaging rather than content.
Vertical 9:16 by default. Respect the safe zones — the top and bottom of the frame get covered by interface, so nothing important lives there. Captions sit in the middle third.
The hook frame matters separately from the hook: the still that shows before playback needs to work as an image on its own. Design it deliberately rather than accepting whatever frame one happens to be.
Deliver in the sizes the client actually needs — feed, story, and a square fallback — rather than making them crop. Sounds trivial. It is the sort of thing that gets you renewed.
Everything so far produces clips. This phase produces sequences, and it is what lets you charge at the top of the ladder rather than the bottom.
Three things carry a multi-shot piece: continuity (the same world across cuts — light direction, time of day, wardrobe), pacing (shot length varying deliberately; everything cut to the same rhythm reads as a slideshow), and escalation (each shot needs to earn the next one).
Work from a shotlist, never from a paragraph. Eight numbered scenes, each with its own prompt, generated separately and assembled. This is the biggest single quality jump available to you and it is expanded as Module 04 later.
Dialogue is the advanced case: keep lines short, one thought per shot, and cut on the beat rather than mid-sentence.
The last phase is about removing yourself from the parts of the work that do not need you.
Cloning first: a single portrait plus an audio file produces a lip-synced presenter, which means one saved face can deliver unlimited scripts. Build two or three house presenters you own outright and you never negotiate with a creator again.
Then automation. Once the craft is repeatable, the clicking is the bottleneck, and the studio can be operated by an AI assistant through its MCP connection rather than by hand — covered properly in Chapter 09.
Clone yourself freely. Clone anyone else only with written permission, and never a real customer, a real founder or a public figure without it. Voice and likeness are legally protected in most places you will be selling, and "the AI made it" has never once worked as a defence.
You can make the thing. Now: who buys it, what it costs, what you send them, and how you keep them past month three.
Before you spend another twenty minutes, here is what you are not being asked to do.
And here is the part the sales pages leave out.
This is not passive. A studio is not a slot machine. You will do roughly ninety minutes of real work per client per month, and the first month of any new client is the hardest because you are learning their product. If you want money to arrive while you sleep without ever having built the thing that makes it, close this and go and be disappointed somewhere else.
Brands are paying $1,500 to $3,000 for AI videos that take a few hours to make.[1]
That is the rate right now — see the artefact below — and it is what businesses are actively paying this week to people who can deliver at the quality bar. Not a forecast. Not a promise.
Sit with that for a second, because it contains both the opportunity and the trap.
The rate above is the ceiling for someone with a portfolio, a reputation and a referral. You will not start there, but you start closer to it than you think — because there are two different ways to be paid, and they behave completely differently.
Per project: $1,000 to $1,500. One campaign, one set of deliverables, one invoice. This is the bottom of the real market band, which is exactly where a newcomer with good work but no reputation belongs. It is the fastest money in this book and the easiest first yes, because a project has an end and nobody is committing to anything.
Per month: $500. Less per hour, and better in every other way. A retainer is predictable, it compounds, month two takes half the work of month one, and twenty of them is a business rather than a run of good luck. You are trading rate for reliability, knowingly.
Do both. Land the project, deliver it well, then convert it: "I can keep this going for five hundred a month." Roughly a third say yes, and a third is all you need.
Go lower and you are competing with every freelancer on every marketplace on earth, all of whom are cheaper than you and always will be. Go much higher on a monthly commitment and you cross the line where a business needs three quotes, a meeting, a procurement conversation and a decision from someone who is not in the room.
$500 sits in the gap. It is small enough that one person can approve it in about ninety seconds on their phone. It is large enough that twenty of them is a living.
Generation cost assumes roughly 15 credits per short video and 3 per still at CR8OR's credit rate. Your real number moves with the models you pick.
Projects pay better per hour and stop when they stop. Retainers pay less per hour and arrive whether or not you found a new client this month.
Twenty retained clients is $10,000 a month against roughly $820 in running costs. That is not a fantasy figure — it is twenty small businesses, which is fewer than the number of independent shops on most high streets.
Reading "$1,500 to $3,000" and opening at $3,000 with nothing to show. People do, they get silence for six weeks, and they quit believing the market was a lie. The market was not a lie — they were a stranger. Start at the bottom of the band, get eight brands who will vouch for you, and then charge the top of it. The order matters more than the number.
Four movements. They repeat every month, forever, per client.
That is the entire mechanism. Everything else in this document is detail underneath those four lines.
You are not selling "AI content". Nobody buys that. You sell four specific, named formats that a brand already knows it needs.
The product, shot beautifully, in motion. Rotating on a surface. Being poured, opened, worn, used. Lit like a commercial. This is the easiest format to make and the one clients react to fastest, because most small brands have never had anything better than a phone photo on a kitchen worktop.
A person to camera, holding the product, talking like a customer rather than a brand. This is the format that carries paid social. You generate the presenter, generate the voice, script it around one problem and one payoff, and keep it under twenty seconds.
The before-and-after. The "I did not expect this to work". Slower, less polished on purpose, shot as if filmed at home. These convert because they look like the thing the viewer was already scrolling past.
The one that looks expensive. Motion graphics, the offer on screen, a hook in the first second. Used for retargeting and for the client's own pride — this is the piece they show their business partner, and it is why they renew.
Meta's delivery system needs a large pool of creative to find a winner — the working assumption among performance marketers is fifty-plus variations before you reliably surface one. Almost no small brand supplies anywhere near that. They make four ads, run them until fatigue, and conclude that paid social does not work for them. You are not selling better taste than their designer. You are selling enough shots on goal, which is a thing they can actually feel in their numbers.
Copy this. Word for word if you like. A productised spec is what stops the conversation becoming a negotiation.
Notice what is not on that list: strategy calls, reporting, account management, revisions without limit. Every one of those is how a $500 client quietly becomes a $500 job. Hold the line and the maths in chapter three stays true.
Do not squirm. The answer is yes, and here is how you say it.
"Yes — it's AI-generated, directed and edited by me. That's exactly why you get thirty pieces for four hundred quid instead of two for two grand. If you want a human film crew I'll refer you to one, and it'll be about fifteen times the price for a fraction of the volume."
Own it immediately and it stops being an objection. Hedge, and it becomes the only thing they think about.
Here is where almost everyone who tries this quietly goes broke before their third client.
To make the four formats properly you need video generation, image generation, voice, lip-sync, scripting and captions. Bought separately, that looks roughly like this.
| Video generation (Runway / Kling tier) | ~$35/mo |
| Image generation (Midjourney tier) | ~$30/mo |
| Voice / TTS (ElevenLabs tier) | ~$22/mo |
| Talking-head avatar (HeyGen tier) | ~$29/mo |
| Scripting (ChatGPT / Claude Pro) | ~$20/mo |
| Editing & captions | ~$24/mo |
| Before you have earned a penny | ~$160/mo |
Illustrative published pricing at the time of writing. Tiers change constantly — which is rather the point.
Six logins. Six billing dates. Six separate places to paste the same brand brief. And every one of them raises its prices or changes its plan the moment you build a workflow on top of it.
That bill is not the real cost, though. The real cost is that none of those tools remember your client. Every generation starts from nothing. You re-type the brand, the product, the tone, the customer, every time, across six tabs, for every one of your twenty clients. That is the thing that makes this business unscalable — not the money.
CR8OR AI Studio is one login that contains the lot: twelve video models, nine image models, two thousand voices, lip-sync avatars, Claude-powered scripting, captions and a media library. There is no subscription. Billing runs on your API key, so you pay for what you generate and nothing else — no six renewal dates, nothing sitting idle charging you on the fourteenth of a month you did not use it.
It is the only place with a model library this wide running on usage-based billing rather than a monthly plan. Everywhere else, breadth is sold as a tier — and the tiers with the models you actually want run into the hundreds per month before you have made a single video.
The part that matters more than the price is the brand context system. You build a client's profile once — brand, products, photos, tone, customer persona — and every studio on the platform is then already holding it. Open video, it knows the brand. Write a script, it knows the tone. Generate an image, the product shot is already sitting in the reference slot.
That is what turns three clients into twenty. Not the saving. The memory.
Worth being clear about what this does and does not replace. There are excellent communities that will teach you this craft cheaply, and they are worth every penny. But a lesson does not generate anything — you finish it and still go and pay six vendors separately. Learning it and running it are two different bills, and only one of them delivers a client's folder on the first of the month.
Credits roll over and do not expire. Commercial rights are included on every plan, which is the bit you need in order to sell the output at all.
All four clips were generated from prompts inside CR8OR AI Studio. No real shoot, no cast, no location. Total generation time: under two hours.
This is the section that earns the price of this document. Run it in order, do not improvise, and one client takes an hour and a half.
Build or open the brand profile. Brand name, site, tone, colours, logo. Two or three products with real photos. One customer persona — who they are, what annoys them, what they are afraid of. Month one this takes fifteen minutes. Every month after, it takes zero.
Pick five angles. It is always the same five, in every niche, forever:
Write one line per angle. That is your entire creative direction for the month.
Generate in batches, not one at a time. Scripts first — all five angles at once. Then voice. Then video. Then statics last, because by then you know which frames worked.
A workable starting prompt for a product video, which you adapt rather than rewrite:
Cinematic product shot of [PRODUCT] on [SURFACE]. Slow push-in, shallow depth of field, soft key light from camera left, subtle rim light. [MATERIAL] texture visible. Neutral background, no text, no logos, no hands. 9:16 vertical. 5 seconds.
And for a UGC-style ad:
Handheld selfie-style video, [AGE] [GENDER] in a home [ROOM], natural window light, slightly imperfect framing. Holding [PRODUCT], talking directly to camera with genuine enthusiasm. Casual clothing. No studio lighting. 9:16 vertical. 8 seconds.
Build a library of these as you go. Within two months you will have a prompt for every format in every niche you serve, and the build stage collapses to almost nothing.
The bin test. A clip only survives if all four are true:
Expect to bin between a third and a half. That is normal and it is the job. Delivering forty mediocre files instead of twenty good ones is the fastest way to lose a client who cannot articulate why they are unhappy.
One folder. Subfolders by format. Filenames like brand_ugc_problem_01.mp4. One short note
naming
the three you would test first and why. Send it on the same date every month without being chased — the
reliability is worth more to them than any individual video.
Everything in this library is in current circulation — these are the methods being taught right now by the people making the best-looking AI ads on the internet. Each one is normally demonstrated on a different paid platform. All eight run in one studio.
Before you generate a single clip for a client, you build exactly three reference images: a face reference (head and shoulders, neutral expression, even lighting), a full-body reference (same person, whole outfit visible), and a clean product shot (isolated, no background clutter).
Every video that brand ever gets is generated from those three files. That is the whole trick. Presenters stop drifting between clips because you stopped asking the model to reinvent them.
In the studio: generate the two people references in Image Studio, save all three to the brand's library, and pull them into the reference slots on every generation.
Deepen module 01 into a proper sheet: the same face at four angles, three expressions, two outfits. Generate it once, keep it forever. When a client wants their presenter in a kitchen in July and a bathroom in August, you are not rolling the dice — you are re-using a person.
Identity is not only the face. Lock the voice too: pick one voice per presenter and note the ID. A consistent face with a different voice each month reads as wrong even when the viewer cannot say why.
In the studio: Image Studio for the sheet, Speech Studio for the fixed voice, both stored on the brand profile.
Do not invent formats. Find an ad already running in your client's category — the ad libraries are public — and decode its structure: hook type, shot count, pacing, where the product first appears, what the last three seconds do.
Then rebuild that structure around your client's product. You are copying the skeleton, never the skin.
Formats currently worth having in your rotation: the street interview, the unboxing, the two-person reaction, the "I bought this so you don't have to" review, and the silent product demo with text overlay.
In the studio: Analyse Media takes the URL of any video and hands you back a structured prompt describing what is in it — which you then edit to swap in the client's product.
Rebuilding a proven format is ordinary advertising practice. Recreating a specific ad's distinctive creative — its script, its characters, its exact shots — is a copyright and passing-off problem, and it is your client who gets the letter. Steal the structure. Never the execution.
Stop prompting for whole ads. Have Claude write a shotlist first — eight numbered scenes, each with its own prompt, each five to eight seconds. Generate them individually, then assemble.
This is the single biggest quality jump available to you. One long prompt gives a model too many decisions; eight short ones give it eight easy jobs. It also means a failure costs you one scene rather than the entire ad.
In the studio: Script Studio writes the shotlist with brand context attached, then each scene goes to Video Studio as its own generation.
Asking for "realistic" produces the opposite. Realism comes from specifying a medium, not a quality: name the camera, the lens, the film stock, the light source, the time of day.
Then add imperfection deliberately — slightly off-centre framing, a hand entering frame, visible skin texture, a wobble in a handheld shot. Perfect is the tell. The reason so much AI creative reads as fake is that it is too clean, too symmetrical and too evenly lit.
Useful additions: shot on 35mm, overcast window light,
visible pores and flyaway hair, slight handheld drift,
no studio lighting.
Clients notice when the kitchen changes between clips in the same ad. Fix it the same way you fixed the face: generate the location once as a still, then use that still as the reference frame for every shot set there — wide, mid, over-the-shoulder, close on the product.
A brand should have three saved locations, not thirty. Consistency reads as production value.
When one clip in eight is wrong, do not regenerate the ad. Describe the change and edit that clip in place — conversational video editing now handles most single-element fixes without touching the rest of the shot.
This is a cost control as much as a quality one. Regenerating a full ad to fix one hand is how people burn through a month's credits in an afternoon.
In the studio: Gemini Omni editing in Video Studio — describe the change, not the whole video.
Set a hard cap before you start: ten generations per finished ad. Not ten attempts at scene one — ten across the whole thing, which given eight scenes means you get roughly two retries and that is all.
The cap forces you to fix the prompt rather than re-roll the dice, which is the actual skill. Studios that do not set a budget do not lose money on credits; they lose the ninety-minute delivery window, and then the whole thing stops working.
Credit costs vary by model. Cheaper models exist for drafts — block out the shotlist on a fast model, then re-run only the keepers on the expensive one.
Once you are past about ten clients, the bottleneck stops being generation and becomes clicking. Opening tabs, choosing models, filling fields, downloading files, renaming them.
This is where you stop using the studio as a website and start using it as a tool your AI can operate.
CR8OR AI exposes an MCP server — Model Context Protocol, the standard that lets an AI assistant call external tools directly. Connect it to Claude and Claude can generate images, generate video, write scripts and synthesise speech on your behalf, inside a normal conversation.
In Claude, add a custom connector pointing at the CR8OR endpoint:
https://www.cr8or.ai/mcp
Authenticate with your CR8OR token. Claude will then list the available tools — image generation, video generation, script writing, speech — and can call them the same way it calls search.
Instead of forty minutes of clicking, the build stage becomes a sentence.
"Here's the brand profile for Northshore Coffee. Write me five UGC ad scripts using the standard five angles, generate a voiceover for each in a British female voice, then generate a matching 8-second vertical video for each one. Vary the presenter."
Claude writes the scripts, calls the speech tool, calls the video tool, and hands you back the files. You spend your ninety minutes on the two things a machine still cannot do for you: choosing the angle and culling the output.
Do not automate the cull. It is tempting and it is the single fastest way to burn a client. An AI will happily hand you a clip with a melting hand and a garbled label because it has no stake in your reputation. Automate the build, never the gate.
You need twenty clients, ever. Not a funnel. Not an audience. Twenty.
There are communities and marketplaces that will hand you work — post a profile, wait, take what comes. Use them for your first job if you like. Just understand what you are agreeing to: your income exists at the pleasure of whoever runs the board. They set the rates, they own the relationship, and the day they change the rules or close the hub, you have nothing — no client list, no contact details, no way to reach the people who paid you.
That is the same trade as an agency retainer, dressed as a community. A client you found yourself is an asset. A client handed to you is a loan. Twenty relationships you own is the difference between a business and a shift.
Serve one niche until you own it. The same five angles, the same prompts, the same objections — the second client in a niche takes half as long as the first, and the tenth takes a quarter. Fields that work well because the product is visual, the margins are decent and the owner is reachable:
Do not pitch. Build first. Pick a brand, take their product photo from their own site, and make one product video and one UGC ad. Thirty minutes of work. Then send it.
Subject: made you something
Hi [NAME] — I made two ads for [PRODUCT] this morning. No catch, they're attached, use them if you want them.
I build campaign sets like this for brands your size — $1,000 for a one-off set, or $500 a month if you want a steady drip. If the two attached are any good, reply and I'll send the spec. If not, no harm — keep them anyway.
[YOUR NAME]
The reply rate on a message with finished work attached is not comparable to the reply rate on a pitch. You are not asking for their attention, you are handing them something they can use this afternoon.
Ten of these a day. Five days a week. You are not looking for a good conversion rate — you are looking for twenty yeses in your lifetime.
"Great — I'm not replacing them. They make your four hero pieces, I make the thirty variations that let the algorithm find a winner. Different job."
"Some of it does, which is why I bin about half before you see any of it. Tell me which of the two I sent looks off and I'll show you the version that doesn't."
"I can do it smaller. Five hundred is thirty pieces. If you want fifteen, that's three hundred. Same quality, half the volume." Never discount the price. Discount the quantity.
The first $500 is the most expensive customer you will ever acquire. Everything after it is nearly free, which is why the ladder matters more than the entry price.
This is also where chapter three gets paid back. You underpriced on purpose to buy proof; by month six you have eight brands who will vouch for you and a reel of work you are allowed to show. At that point the top of the band stops belonging to other people. Existing clients move up the ladder, project fees go to $2,500 and beyond, and new enquiries start at a number you would not have dared say out loud in week one.
| Project — one campaign set, one invoice | $1,000–$1,500 |
| Starter — 8 videos, 30 statics, monthly | $500/mo |
| Growth — 16 videos, 60 statics, weekly drops | $1,200/mo |
| Full feed — everything, plus running their paid social | $2,500/mo |
Roughly a third of clients who stay six months will move up a rung if you ask. You do not need a sales process for this — you need to notice when a client starts asking what to do with the content, because that question is the upgrade signal.
Above the top rung, the conversation changes entirely. At that point they are not buying content, they are buying infrastructure — ad accounts, email, SMS, the whole owned system. That is where CR8OR's Ad System sits, and where your $500 studio quietly becomes something considerably larger.
Four things that are true and that nobody selling you a system will tell you.
Not month one. Month one they are delighted. Month three is when the novelty has gone and they look at the folder and think: this all looks a bit the same. The fix is to change the angle mix every month and to keep culling harder than feels necessary.
The best-performing feeds are not fully synthetic. They mix generated work with real footage, real founders, real customers. Position yourself as the volume layer in a real brand's content, not as a replacement for everything human in it. You will keep clients twice as long.
Disclosure requirements and labelling rules for synthetic media are tightening across every major platform, and they will change again after this is written. Check the current rules before you deliver, tell your client plainly what is generated, and never let a brand run something as a real customer testimonial when it is not one. That is not caution — that is the difference between a business and a liability.
There is no version of this where twenty brands find you. Ten outreach messages a day for a month is the entire growth strategy, and most people who read this will send four and then stop. Your competition is not other studios. It is your own week two.
Open the studio. Pick one niche. Build a brand profile for a real brand you do not work with, and produce one full batch of eight videos and thirty statics for them. This is your portfolio, your practice run and your first outreach asset, all at once.
Do it again for a second brand and time yourself. Write down every prompt that worked. Fix the step that took longest. By the end of the week you should be under two hours per client.
Ten outreach messages a day, every one with finished work attached. Track replies in a spreadsheet. Do not change the message until you have sent fifty — you cannot read a signal from twelve.
Close your first one or two as projects — a project is an easier yes than a commitment. Deliver early, then ask the conversion question: "I can keep this going for five hundred a month." Put the delivery date in the calendar as a recurring event. Then connect the MCP server and start compressing the ninety minutes.
At day thirty you have one client or you have nothing. There is no partial version of this — and one client, run properly, becomes twenty by repetition rather than by insight.
Everything in this document runs on one platform. Twelve video models, nine image models, two thousand voices, lip-sync avatars, Claude-powered scripting and a brand context system that remembers every client you take on.
The part that matters for a studio charging $500 a month: there is nothing to subscribe to. Billing runs on your API key, so a quiet month costs you nothing and a busy one costs you what you used. Every comparable platform sells breadth as a monthly tier, and the tiers that carry the models you actually want run into the hundreds — before a single video exists.
And if you get to the top of the ladder — the client who wants the ads run, the email, the SMS, the whole owned system rather than just the creative — that is the CR8OR AI Ad System, and it is a conversation worth having.