Faceless Channel Cost: What Each AI Video Costs to Make

A ten-minute faceless video used to mean two invoices: an editor and a narrator. Using the rate ranges published by Vidico and the GVAA figures summarised by VoiceCrafters, those two lines together land somewhere between $800 and $1,700 per video. The same video built from API calls has a machine cost in the tens of dollars, and most of that goes on visuals.

Most people budgeting a faceless channel fixate on the cheap lines (script, voice) and wave through the one that decides the budget: seconds of generated video. Below is a cost model you can rebuild in a spreadsheet, applied to a 30-second short, a 60-second short and a 600-second long-form video, with every price dated and linked.

The five lines that make up a faceless video's cost

Script. Tokens from a language model. Priced per million tokens.
Voice. Text-to-speech, priced per character of narration.
Moving visuals. Generated video clips, priced per second of output.
Still visuals. Generated images used as slides, backgrounds or thumbnails, priced per image.
Rerolls. The takes you throw away. Nobody publishes this number, so you have to set it yourself.

Assembly (cutting, captions, music, export) is paid in your time, so it stays out of the machine cost and comes back in the comparison with hired help.

The formula the rest of this article uses:

Cost per video = script + voice + (video seconds x rate per second + stills x rate per image) x reroll factor

The reroll factor multiplies visuals only, because a voice line or a script rarely needs more than one pass to be usable, while a clip with a warped hand does. I use 1.5 as a planning assumption, meaning one extra take for every two you keep. Track your own ratio for a month and replace it.

Script: the line you can stop worrying about

Anthropic's pricing page gives a useful conversion: one token is roughly four characters, or 0.75 words, in English. The same page lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, and Claude Haiku 4.5 at $1 and $5.

To size the scripts I assume a narration pace of about 150 words a minute. That is a planning figure, and your voice and niche will move it, so time one of your own voiceovers and adjust.

30 seconds: about 75 words, or 100 output tokens.
60 seconds: about 150 words, or 200 output tokens.
600 seconds: about 1,500 words, or 2,000 output tokens.

A 600-second script on Sonnet 5.5 is 2,000 output tokens x $10 per million, which is $0.02. Add a 5,000-token prompt with your outline and research notes ($0.01) and three rounds of revision, and you are still under $0.10. The Batch API halves both rates if you generate a week of scripts overnight.

Pick the model that writes the best script for your niche and ignore its price, since the spread is cents per video.

Voice: cheap per video, worth checking per character

Narration length converts to characters with the same Anthropic rule of thumb: one word is about 5.3 characters (four characters per token divided by 0.75 words per token). Rounded:

30 seconds: about 400 characters.
60 seconds: about 800 characters.
600 seconds: about 8,000 characters.

On Aitachyon, ElevenLabs voiceover costs $0.043 to $0.086 per 450 characters at the time of writing, depending on the voice model. That gives:

30 seconds: one 450-character block, $0.043 to $0.086.
60 seconds: two blocks, $0.086 to $0.17.
600 seconds: about 18 blocks, $0.77 to $1.55.

Buying direct is cheaper per character. ElevenLabs' own API page lists v3 and Multilingual v2 at $0.08 per 1,000 characters and Flash/Turbo at $0.04, so the 8,000-character narration costs $0.32 to $0.64 there. A third-party breakdown from Puter quotes $0.05 and $0.10 per 1,000 characters instead. When the vendor page and a third party disagree, trust the vendor. If voice is a meaningful share of your spend, which is rare for a faceless channel, a direct account is the cheaper route. For choosing between voice models on quality rather than price, the ElevenLabs voiceover cost breakdown goes further.

Visuals: the line that decides your budget

Here is where faceless channels diverge. You can fill the screen three ways, and the choice changes your cost by an order of magnitude.

Mode A: stills with motion added in the edit

Generated images, panned and zoomed in your editor. Prices at the time of writing: Seedream 4.0 at $0.057 per image, FLUX.2 [pro] at $0.057 to $0.086, Nano Banana at $0.13, and gpt-image-2 at $0.10 to $0.40. For reference, Google's direct price for Nano Banana 2 runs from about $0.045 at 0.5K to $0.151 at 4K, and fal lists the original Nano Banana at $0.0398 per image.

Mode B: generated video for every second

The most watchable and the most expensive. Per-second prices at the time of writing: Hailuo 02 at $0.085, Kling v3 at $0.16 silent and $0.32 with native audio, Veo 3.1 Fast from $0.19, Wan 2.7 at $0.19, and Seedance 2.5 from $0.20 at 480p and $0.44 at 720p.

Mode C: hybrid

Generated video where motion earns its cost (the hook, a demonstration, a transition), stills everywhere else. Most channels that last end up here, because it is the only mode where a ten-minute upload stays in the tens of dollars and still moves on screen.

A faceless channel rarely needs native audio in its clips, since narration and music go on top, so silent per-second rates are the relevant ones. The model-by-model video API pricing guide covers the resolution and audio trade-offs in more detail.

The three videos, priced line by line

All prices below are Aitachyon's at the time of writing. The reroll factor of 1.5 applies to visuals only. Stills are budgeted at one image per 5 seconds for shorts and one per 6 seconds for long-form, which is a pacing assumption you should replace with your own.

30-second short

Script: under $0.01.
Voice: $0.043 to $0.086.
Mode A (6 Seedream stills): 6 x $0.057 = $0.34, x 1.5 = $0.51. Total about $0.56 to $0.60.
Mode B on Hailuo 02: 30 x $0.085 = $2.55, x 1.5 = $3.83. Total about $3.87 to $3.92.
Mode B on Kling v3 silent: 30 x $0.16 = $4.80, x 1.5 = $7.20. Total about $7.24 to $7.29.

60-second short, hybrid

Script: under $0.01.
Voice: $0.086 to $0.17.
Video: a 5-second hook plus two 5-second inserts on Hailuo 02, 15 x $0.085 = $1.28.
Stills: the remaining 45 seconds at one per 5 seconds, 9 x $0.057 = $0.51.
Visuals with rerolls: ($1.28 + $0.51) x 1.5 = $2.69.
Total: about $2.78 to $2.87.

600-second long-form, hybrid

Script: under $0.10 with revisions.
Voice: $0.77 to $1.55.
Video: 60 seconds of motion (10 percent of runtime) on Hailuo 02, 60 x $0.085 = $5.10.
Stills: 540 seconds at one per 6 seconds, 90 x $0.057 = $5.13.
Visuals with rerolls: ($5.10 + $5.13) x 1.5 = $15.35.
Total: about $16.20 to $17.00.

For contrast, the same 600 seconds as pure Mode B on Hailuo 02 is 600 x $0.085 x 1.5 = $76.50 in visuals alone, and on Kling v3 silent it is $144. Full-motion long-form is the one configuration where AI video stops being cheap, so decide deliberately which seconds deserve to move.

Batches: what a week and a month actually cost

The planning number is your monthly run rate. Three realistic batches:

A week of daily shorts (7 x 60-second hybrid): 7 x about $2.85 = about $20.
20 hook variants to test openings (20 x 5 seconds): 100 seconds on Hailuo 02 is $8.50; on Kling v3 with native audio it is 100 x $0.32 = $32. Testing hooks cheaply first and upgrading the winner is the sensible order.
A month for a two-format channel (4 long-form + 20 shorts): 4 x $17 + 20 x $2.85 = $68 + $57 = about $125.

Thumbnails add a little. Fifty thumbnail candidates on Seedream 4.0 cost 50 x $0.057 = $2.85, or $5 to $20 on gpt-image-2 depending on quality. Generate many, keep two, test them.

The way you pay for these batches matters as much as the unit price. Monthly credit plans that reset tend to punish uneven publishing schedules. The comparison of credits against pay-per-call billing works through that with examples.

The human benchmark, with the same arithmetic

To keep the comparison fair, here is what the equivalent work costs when you hire it, using published rate ranges.

Editing

Vidico puts freelance editors at $25 to $150+ per hour, and per finished minute at $20 to $100 for long-form YouTube and $10 to $50 for TikTok and Reels. Per project, it lists $300 to $1,500 for a YouTube video and $50 to $400 for a TikTok or Reel. A full-time in-house editor runs $51,000 to $93,000 a year.

Narration

VoiceCrafters' summary of the GVAA rate guide lists non-broadcast narration at $350 to $450 for 1 to 2 finished minutes, $600 to $700 for 5 to 10 minutes, and $700 to $800 for 10 to 15 minutes. YouTube work with paid placement falls under digital broadcast, which the same guide prices at $1,000 to $1,250 for a three-month national usage term.

Side by side

600-second video, hired: editing at $20 to $100 per minute is $200 to $1,000, plus narration at $600 to $700. Total $800 to $1,700.
600-second video, API calls: about $16 to $17, plus your assembly time.
60-second short, hired: editing at $50 to $400 per project, plus narration at $350 to $450. Total $400 to $850.
60-second short, API calls: about $2.85, plus your assembly time.
The month above (4 long + 20 shorts), hired: 4 x $800 to $1,700 plus 20 x $400 to $850, which is $11,200 to $23,800.

Two honest caveats. First, the hired figures buy judgement: an experienced editor fixes pacing and story problems that no model will flag. Second, your time is not free. If assembling a ten-minute video takes you three hours, price those hours at whatever your time is worth and add them to the AI column. The comparison still favours generation by a wide margin at volume, and it narrows fast if you are making one video a month and value your evenings. The broader trade-offs are laid out in the guide to running a faceless YouTube channel with AI, and before you plan revenue on top of these costs, read what YouTube's rules say about monetizing AI content.

Buying models direct versus through one account

Every model above can be bought from its maker or from a reseller, often at a lower list price per unit. Examples as of September 2026:

Google lists Veo 3.1 Fast at $0.10 per second at 720p with audio, $0.12 at 1080p, and Veo 3.1 Lite at $0.05 per second at 720p. Veo 3.1 Standard is $0.40 per second.
fal lists Kling v3 Pro text-to-video and image-to-video at $0.14 per second, and Kling v2.5 turbo at $0.07.
Costbench shows Kling API pricing across providers ranging from $0.056 to $0.42 per second, and Atlas Cloud lists Kling V3 Turbo from about $0.095 per second.
If your format uses a talking avatar as narrator, Runware prices Kling AI Avatar 2.0 Standard per generation, from about $0.199 to $0.868, without a per-second rate on the page.

For one model used at high volume, direct usually wins on unit price and you should buy it that way. A channel that mixes a cheap model for hooks, an image model for stills, another for thumbnails and a voice model also pays in five accounts, five keys and five invoices, a cost that never shows up in a per-second comparison. Decide per line: if a single model is more than half of your monthly spend, price it direct.

Running it from code instead of a browser tab

These numbers assume volume, and volume is where clicking through web interfaces breaks down. A faceless channel is a pipeline, and every step except taste can run from a script or an agent:

Write the shot list as data. Each line of narration gets a shot type (still or clip), a prompt and a duration. This is where you enforce the 10 percent motion budget from the long-form example.
Estimate before generating. Multiply the shot list by the machine-readable price grid (Aitachyon publishes it at /api/pricing) and refuse to run if the estimate exceeds your per-video ceiling.
Generate in one pass. Call the API with an Authorization: Bearer ait_... header, or let an agent do it. In Claude Code the hosted MCP server is one command: claude mcp add --transport http aitachyon https://aitachyon.com/api/mcp --header "Authorization: Bearer ait_...". After that, you can hand the agent the shot list and it calls the models itself. The Claude Code and MCP walkthrough shows the full loop.
Store refs, not files. Each output gets a stable ref (img_, scn_, vo_) that your assembly script fetches with GET /api/generations/{ref}. Rerolls become a matter of regenerating one ref.
Reconcile actual against estimate. Every job is itemised with its model and cost, so you can compute your real reroll factor per model and feed it back into step 2.

Two safety details matter once an agent is spending money for you. Give each pipeline its own API key, so spend is attributed per channel and an alert fires when a key spends unusually fast. And when a render fails, it is refunded automatically, so a flaky batch does not quietly inflate your unit cost. At a few hundred clips a week, the rate limits, retries and queueing become their own topic, covered in producing hundreds of clips safely. Setup details for the API and MCP server are on the developer quick start.

A checklist for cutting cost without cutting quality

Set a per-video ceiling (for example $3 per short, $20 per long-form) and make your pipeline enforce it.
Budget motion as a percentage of runtime. Start at 10 percent for long-form and raise it only where retention data says motion helps.
Test on the cheapest model that looks acceptable. Run hook variants on Hailuo 02 at $0.085 per second, then regenerate only the winners on a more expensive model.
Buy silent clips when narration and music go on top. Kling v3 silent is half the price of Kling v3 with audio at the time of writing.
Use the lowest resolution your platform needs. Seedance 2.5 more than doubles from 480p to 720p on the current price list.
Measure your reroll factor per model. A cheaper model that needs three takes can cost more than a pricier one that needs one.
Price your hours. If assembly time grows faster than output, that is the line to automate next.

FAQ

How much does it cost to run a faceless YouTube channel with AI per month?

With the prices above, four ten-minute hybrid videos and twenty 60-second shorts come to about $125 a month in model calls, before your own time. Pure full-motion long-form multiplies the long-form line by four or more, so the motion budget is the setting that matters most.

What is the most expensive part of an AI faceless video?

Generated video, by a wide margin. A ten-minute script costs cents and its narration under $2, while 600 seconds of full-motion video costs $76.50 or more on Hailuo 02 once rerolls are counted.

Is AI voiceover cheaper than hiring a voice actor?

For volume narration, yes. A 5 to 10 minute non-broadcast narration is listed at $600 to $700 in the GVAA figures summarised by VoiceCrafters, while 8,000 characters of ElevenLabs narration costs under $2 through an API.

How much does a video editor charge for a YouTube video?

Vidico lists $300 to $1,500 per YouTube video on a per-project basis, or $20 to $100 per finished minute for long-form. Short-form TikTok and Reels projects run $50 to $400.

Can I automate a faceless channel end to end?

Script, narration, visuals and cost tracking can all run from code or an agent. Topic choice, the final edit pass and the decision to publish are still better done by a person, and platform rules on AI content should be checked before you scale.

Sources

If you want to run this cost model against real invoices, Aitachyon puts every model priced above behind one prepaid balance and one API key, usable from the web studio, plain HTTP or the MCP server in Claude Code, with each job itemised by model and cost and failed renders refunded. The full price list is on the models page.

Faceless Channel Cost: What Each AI Video Costs to Make