Video sales letter with AI: model a VSL that already sells, in any language, without filming

A video sales letter is a sales page read aloud. The original format, the one that built the direct response industry online, is a voice over plain text appearing on screen, sentence by sentence, for anywhere between eight and forty five minutes. No presenter, no studio, no b-roll of smiling people. Marketers call it the ugly VSL and it has beaten the polished, filmed version in more split tests than anyone has bothered to count, because the viewer is not there for production value. They are there for the argument.

That detail changes what producing a VSL actually is. Take the spoken rate most narrators settle into, around 150 words a minute, and the arithmetic is simple: a 10 minute VSL is about 1,500 words, a 20 minute one is 3,000, and the long 30 minute version for cold traffic is 4,500 words. Every one of those words is a decision about the argument, and none of them requires a camera. The VSL is the only video in the marketing stack where the words are the entire product and the picture only has to keep the eye busy while the ear does the buying.

That is precisely the shape of work an AI video pipeline is good at. FalconVid's specialists write from a structure, narrate in an ultra realistic premium voice or your own cloned voice, put the words on screen in a karaoke style caption with more than 15 presets, generate scenes that match the noun being said, and deliver a finished file in up to 30 minutes. What used to be a three week project across three freelancers becomes a project on your calendar, and the rest of this guide is about making sure the words that go into it are the ones that sell.

• The classic VSL is a voice over text on screen, 8 to 45 minutes, no camera
• At 150 words a minute: 10 min = 1,500 words, 20 min = 3,000, 30 min = 4,500
• The words are the product; the picture only keeps the eye busy while the ear buys

Every VSL that has ever converted follows the same six part skeleton, and the skeleton is what a modeled VSL copies. The lead is the first 30 to 90 seconds: a promise or a pattern interrupt that earns the next minute. The problem section names the pain in the viewer's own words and makes the cost of leaving it alone concrete. The mechanism explains why the pain exists and why the usual fixes fail, which is the section that separates a VSL from an ad. Proof follows, in numbers, cases and testimonials. Then the offer: what they get, what it would cost elsewhere, the price, the terms. Then the close, which repeats the promise, removes the last objection and tells them exactly what to click.

Length follows traffic temperature, not preference. The market convention that has held for years is that cold traffic, people who have never heard of you, needs the long version, 20 to 45 minutes, because the mechanism and the proof have to do all the work. Warm traffic, retargeting or an email list, converts on 5 to 12 minutes. The mini VSL that sits inside a paid ad is 3 to 8 minutes and exists mostly to sell the click to the long one. The structure is identical at every length; what changes is how many proofs and how many objections you have room to handle.

Modeling means copying the skeleton, never the flesh. A reference VSL from your market tells you the order of the sections, where the proof goes, how long the lead runs and where the price appears. Your words, your mechanism, your proof and your offer fill the skeleton. The distinction matters commercially and legally: copying someone's script is a copyright problem and a positioning problem at once, while copying the tested order of an argument is just how direct response has always worked. The guide to writing a script for an AI avatar shows the dose logic that keeps a narrated argument from sagging; in a VSL that logic runs for the whole piece.

• Six parts: lead, problem, mechanism, proof, offer, close, always in that order
• Cold traffic 20 to 45 min, warm traffic 5 to 12, mini VSL inside the ad 3 to 8
• Model the skeleton, never the flesh: order and proportions, not sentences

The manual VSL is a three vendor project. The copywriter comes first, and the ranges quoted on freelance marketplaces for a full VSL script run from about US$ 500 for a junior writer to US$ 5,000 and beyond for someone with a track record in your niche, with royalty deals on top for the ones who have earned them. The voice comes second: a professional voice actor charges in the range of US$ 100 to 400 for a finished 20 minutes, more for a rush, more again for revisions after the copy changes. The editor comes third, at US$ 300 to 2,000 for the text-on-screen assembly, music bed, captions and export, depending on whether they are also sourcing footage.

The number that hurts is not the total, which lands somewhere between US$ 900 and US$ 7,000 for one VSL in one language. It is the calendar. Three vendors in sequence means three queues, and a script revision after the voice is recorded restarts two of them. The realistic timeline from brief to a published first version is three to six weeks, and that first version is one guess at the lead, one guess at the offer, one language. The second guess costs most of the first all over again, because the voice actor and the editor bill by the pass, not by the project.

In an automated pipeline the same VSL is a calendar item and a credit count. On today's ruler a 10 minute narrated piece costs 1,524 credits in the economy mode, a 12 minute one 1,731, a 20 minute VSL 2,557 and a 30 minute cold traffic version 3,593. The balanced mode, with richer scenes, is 3,217 credits for 8 minutes and 4,597 for 12; premium, with Veo 3 audio and Seedance Pro scenes, is 10,911 for 8 minutes and 16,131 for 12. None of those numbers changes when you revise the lead, because the second version is simply a second run, ready in up to 30 minutes instead of a second invoice. For a VSL the number that matters is how cheap the second guess has become.

• Manual: copywriter US$ 500 to 5,000, voice US$ 100 to 400 per 20 finished minutes, editor US$ 300 to 2,000
• Three vendors in sequence: 3 to 6 weeks to a first version, and one revision restarts two queues
• Automated, today's ruler: 10 min 1,524 credits, 20 min 2,557, 30 min 3,593 in economy; the second version is a second run, not a second invoice

The most common mistake with a winning VSL is to translate it. The words move to Spanish or Portuguese and the conversion rate falls off a cliff, and the owner concludes that the market is bad. The market is fine; the VSL was translated when it needed to be modeled. The skeleton survives the border. Almost everything inside it does not: the currency and the price anchor, the proof, which has to come from people the new audience recognizes as their own, the objections, which differ by market, the guarantee wording, which is regulated differently, and even the length of the lead, because a promise that reads as confident in one culture reads as a scam in another.

This is where the pipeline's multilingual design becomes the whole business case. FalconVid narrates in 63 languages with premium ultra realistic voices or your cloned voice, and a project can be duplicated into another language paying only the difference, so the Spanish and Portuguese versions of a US VSL are not three productions but one production and two adaptations. The structure is replicated, the market specific sections are rewritten, the voice stays yours. The guide to cloning your voice with AI covers what a clone does well and where it still needs a human ear; for a VSL the point is that the founder's voice sells in a language the founder does not speak.

The commercial arithmetic is the reason to bother. A VSL that converts at a healthy rate in English is a proven argument, and proven arguments are the scarcest asset in marketing. The Spanish speaking market alone is over 500 million people, most of them served by offers that were translated rather than modeled, and the Portuguese market adds a Brazil where video is the default way to buy online. Duplicating a proven VSL into those two languages, with the market specific sections rewritten, is the cheapest new revenue a direct response business can add, and the reason the AI video automation hub treats the VSL as a product line rather than a one off video.

• Translation moves the words; modeling keeps the skeleton and rewrites the market specific sections
• What changes: currency and anchor, proof, objections, guarantee wording, length of the lead
• 63 languages, duplicate paying only the difference, the founder's cloned voice in a language the founder does not speak

Retention curves of VSLs all look the same: the steepest drop is in the first 60 to 90 seconds, and whatever share of viewers survives the lead is roughly the share that hears the offer. That makes the lead the highest leverage 90 seconds in the whole funnel, and it is also the section with the most possible variants: a question, a contrarian claim, a story opening, a demonstration, a direct promise. In the manual model you get one, because the voice actor and the editor have already been paid. In the automated model you get as many as you want to test, because each one is a run.

The practical routine is five leads on the same body. Write five openings of 60 to 90 seconds, keep the problem, mechanism, proof, offer and close identical, and produce five versions. On today's ruler five 10 minute VSLs in the economy mode are 7,620 credits, which fits inside a Starter month with 15,000 credits and room left for the long version. Run them against the same traffic source, read the hold at 90 seconds and the click rate on the offer, keep the winner, and write five more leads against it. Two rounds of that are ten tested openings in a month, which is more lead testing than most offers see in their lifetime.

The same logic applies to the offer section and to the close, and to the retargeting cut. A 20 minute cold VSL yields a 6 minute warm version by keeping the lead, the strongest proof, the offer and the close, and a 60 second Shorts 9:16 cut for the ad in front of it by keeping the lead alone. FalconVid produces the 9:16 cut from the same project, so the ad, the long VSL and the retargeting version share one structure and one voice, and a test result in one of them tells you something about all three.

• The steepest drop is the first 60 to 90 seconds; the survivors of the lead are roughly the ones who hear the offer
• Five leads on one body: 5 × 1,524 = 7,620 credits at 10 minutes in economy, inside a Starter month
• One structure, three cuts: 20 min cold, 6 min warm, 60 s vertical for the ad

A VSL has fewer places to look fake than a documentary, which is good news, and the places it has are all in the voice. A monotone narration reading 3,000 words at a constant pitch is the single loudest tell of machine production and the single fastest way to lose the lead, because a sales argument lives on emphasis: the pause before the price, the slowed line where the mechanism lands, the lift when the promise is repeated in the close. Cheap text to speech has none of that. The premium voices and the cloned voice in FalconVid carry emphasis and pacing, and the Studio lets you hear the first version and tighten the opening before a single viewer does.

The second tell is the picture. The classic text on screen VSL avoids it by having almost no picture, which is a legitimate choice; the modern VSL that adds scenes has to add ones that match the noun being said, not generic stock of hands typing. The checklist for making AI video look less like AI applies in full: cut where the argument turns, match the image to the exact claim, keep the caption style consistent, and never let a smiling stock model appear during the problem section. FalconVid's scene generation follows the script noun by noun, and the DNA of the project keeps the caption style, the voice and the visual identity identical across every version and every language.

The third tell is the one nobody plans for: consistency across versions. When the English VSL, the Spanish one and the five lead variants each come from a different freelancer, the brand has six voices and six caption styles, and a viewer who saw the ad and then the long version feels the seam. A pipeline that produces all of them from one project has no seam, which is a quiet, measurable lift in the click from ad to VSL.

• The loudest tell is a monotone reading 3,000 words; a sales argument lives on emphasis and pause
• Scenes match the noun being said, or the VSL stays text on screen; never stock smiles during the problem
• Six freelancers produce six voices and six styles; one project produces one brand across every version

Inside FalconVid the VSL is one of the formats the same machine that runs your channels also produces. You paste a reference VSL from your market, and the DNA replicates the selling structure: hook, pain, offer, in the order and proportions the reference proved, with your mechanism, your proof and your terms filling it. The specialists work in parallel, a researcher grounding the claims, a writer producing the script from the structure, a narrator in one of 63 languages with a premium ultra realistic voice or your cloned voice, an editor generating the scenes and the karaoke style captions, a sound designer laying the music bed and mastering the audio, and a finished VSL is ready in up to 30 minutes, with 2 simultaneous generations on the Starter, 5 on the Pro, 10 on the Business, 25 on the Agency and 50 on the Scale.

The pieces around the VSL come from the same project. UGC style product testimonials, delivered by ultra realistic avatars with clean lip sync, make the social proof section and the ads that run in front of the VSL; the guide to AI avatar UGC ads covers what they cost and where they convert. Shorts 9:16 cut from the VSL become the ad creative. The Studio lets you watch the first version, shorten the lead, swap a scene or the music, or open the timeline, instead of regenerating. And when the version is right, the same project duplicates into Spanish, Portuguese or any of the other languages paying only the difference, with the market specific sections rewritten and the structure untouched.

Publishing is a choice, not a task. A VSL usually lives unlisted on YouTube and embedded on the sales page, or on Rumble for offers the YouTube ad policy dislikes, and FalconVid publishes to YouTube, Instagram, TikTok, Rumble and Facebook on the scheduled date, with automatic comment replies in the commenter's language for the versions that run public. Every creation feature is on every plan; what changes as you go up is volume, channels and simultaneous generations, plus the dedicated server and the Senior AI Analyst from the Pro upward, who reads the account and writes to you every two days with the next move, which for a VSL business is which lead won and what to test next.

• Paste a reference: the DNA replicates hook, pain and offer with your mechanism, proof and terms
• Narrated in your cloned voice or a premium voice in 63 languages, ready in up to 30 minutes
• UGC testimonials, 9:16 ad cuts and the Studio for fixes come from the same project

A VSL is advertising, and advertising law does not care that a machine wrote it. In the United States the FTC requires that every claim be substantiated before it is made and, since the 2023 update of the Endorsement Guides, that testimonials reflect what a typical customer can expect or clearly disclose that they do not, with disclosures that are impossible to miss, not buried in a description. In Brazil the Consumer Code's article 37 makes misleading advertising illegal and CONAR enforces the industry's own rules on top. In Spain the General Advertising Law and the Unfair Competition Law prohibit misleading claims and unverifiable comparisons. The structure travels between markets; the specific claims have to be checked in each one, which is exactly the section you rewrite when you model instead of translate.

On today's ruler a 12 minute piece costs 1,731 credits in the economy mode, 4,597 in balanced and 16,131 in premium, and the VSL lengths fall on the same curve: 8 minutes 1,317, 10 minutes 1,524, 20 minutes 2,557 and 30 minutes 3,593 in economy, 3,217 for 8 minutes in balanced, 10,911 for 8 minutes in premium. The Starter includes 15,000 credits, 1 channel and 2 simultaneous generations: one 20 minute cold VSL plus five 10 minute lead tests is 10,177 credits, with room for the warm cut. The Pro includes 30,000 credits, 5 channels and 5 at once, plus the dedicated server and the Senior AI Analyst, which is the plan for an offer running in two or three languages at the same time.

The Business includes 95,000 credits, 10 channels and 10 at once, the plan of an agency running VSLs for several clients. The Agency includes 190,000 credits, 25 channels and 25 at once, and the Scale 320,000 credits, 50 channels and 50 at once. Nobody runs a month in one mode; you mix, and a balanced version of the winning lead next to economy versions of the tests is the honest way to spend a Starter. The free plan produces 4 videos of up to 3 minutes every 30 days with no card, enough to hear your cloned voice deliver a mini VSL lead before you commit, and the annual plan charges 10 months instead of 12.

• FTC substantiation and 2023 Endorsement Guides, Brazil's article 37 and CONAR, Spain's General Advertising Law: claims are checked per market
• Today's ruler: 10 min 1,524, 20 min 2,557, 30 min 3,593 credits in economy; 12 min 1,731 / 4,597 / 16,131 by mode
• Starter: one 20 min VSL plus five lead tests = 10,177 of 15,000 credits; Pro adds 5 channels and the Senior AI Analyst