How to Make ChatGPT Write Like a Human: Our Real Before and After
Make ChatGPT write like a human with 5 editing rules, not prompt tricks. Raw draft: 19 banned words per 994. Our 35 published posts: 16, all quotes.
Create
July 31, 2026 · 7 min read
Tools coveredGoogle Veo · Runway · Sora
I wanted to hand Sora 2, Runway and Google Veo the same ten-second prompt and line the three clips up next to each other. I never got there, because two of the three will not make you a ten-second clip at all. What I found instead is the number none of these vendors prints on their pricing page: what one clip actually costs, which runs from $0.40 to $8.40 for roughly the same ten seconds of video.
This is for anyone deciding where to put their first $20 in AI video. By the end you will know the real per-clip price, the longest clip each one can produce in a single generation, who owns the result, and which of the three is on a countdown clock.
On the record before anything else: I did not generate a single frame for this piece. All three sit behind paid, logged-in accounts, so this is a pricing and documentation teardown built from each vendor's own API reference, pricing page and support docs, checked on 30 July 2026.
The cheapest clip with sound is Google Veo 3.1 Lite at $0.40 for eight seconds. The most expensive is Sora 2 Pro at $8.40 for twelve seconds at 1080p. That is a 21x spread, and none of the three publishes a benchmark that justifies it.
| Model | Longest single clip | Cost of that clip | Audio | Watermark |
|---|---|---|---|---|
| Veo 3.1 Lite (720p) | 8 seconds | $0.40 | Yes | SynthID, always |
| Veo 3.1 Fast (720p) | 8 seconds | $0.80 | Yes | SynthID, always |
| Sora 2 (720p) | 12 seconds | $1.20 | Yes | OpenAI logo + C2PA |
| Runway Gen-4.5 | 10 seconds | $1.20 | Yes | Free plan only |
| Veo 3.1 (1080p) | 8 seconds | $3.20 | Yes | SynthID, always |
| Sora 2 Pro (1080p) | 12 seconds | $8.40 | Yes | OpenAI logo + C2PA |
Those are developer prices, which are the only ones any of these companies quote in plain dollars. Buy the same thing as a consumer through a subscription and the math changes, usually against you.
I priced these three; I did not generate with them. Every figure in this article comes from a vendor's own pricing page, API reference, terms of use or support article, and I have the exact wording for each one in my notes.
A fair generation test would need three paid logins and a card on file, which is not something I do for an article, so I treated the docs as the test subject instead. Reading three pricing systems into one comparable table is exactly the kind of grunt work I once tried to hand to NotebookLM's data tables, and it turns out a human with a calculator is still faster when the units don't match.
Runway is the only one of the three that can produce a ten-second clip in a single generation. This is not a subtlety buried in the fine print, it is right there in the API parameters.
OpenAI's video creation reference is blunt about it: "Clip duration in seconds (allowed values: 4, 8, 12). Defaults to 4 seconds." Google's Veo documentation lists the available lengths as "8 seconds, 6 seconds, 4 seconds", and adds that the higher resolutions are "1080p (8s length only), 4k (8s length only)". Runway's API changelog says "Gen-4.5 supports both text-to-video and image-to-video generation modes, with durations from 2-10 seconds."
So Runway can hit ten exactly. Sora 2 has to jump to twelve, and Veo caps at eight.
Here's the weird part: OpenAI's own docs disagree with OpenAI's own API. The video generation guide claims that "both sora-2 and sora-2-pro support 16- and 20-second generations", while the create reference lists 4, 8 and 12. A developer on OpenAI's forum hit exactly that wall on 16 March 2026, reporting "a 400 error stating that my input of 16 seconds is an invalid integer". If you are budgeting by the second, check the reference and not the guide.
Inside Google Flow, a Veo 3.1 Fast generation costs 20 credits whether the clip runs 4, 6 or 8 seconds. Google's own credit table puts all three lengths in a single cell, "4s, 6s, 8s videos", against one price: 10 credits for Lite, 20 for Fast, and "All users: 100" for a Quality generation, which the table only offers at 8 seconds.
It gets stranger one row down. Gemini Omni Flash, in the same table, is priced by length: 15 credits for 4 seconds, 20 for 6 seconds, 25 for 8 seconds, 30 for 10 seconds. Google prices two of its video models per generation and a third per second inside a single support article, and never mentions that the logic changes.
The plan math lands cleanly. Google AI Pro is $19.99 a month for 1,000 Flow credits, which is exactly ten Veo 3.1 Quality generations, or $2.00 per clip. Through the Gemini API those same eight seconds at Standard quality run $0.40 per second, so $3.20. For a full-length Quality clip, Flow is the cheaper door by $1.20.
That daily allowance is enough to run your own Veo comparison in an afternoon without paying anyone. See for yourself before you subscribe to anything; the exact recipe is at the bottom of this piece.
A Runway credit costs $0.01 in the developer portal and about $0.019 on the cheapest subscription. Same company, same unit, nearly twice the price depending on which page you land on.
Runway's developer docs state it plainly: "Credits can be purchased for $0.01 per credit in the developer portal for an organization." Meanwhile the consumer Standard plan is $12 a month billed annually for 625 credits, which works out to $0.0192 each. On monthly billing it is $15 for the same 625 credits, or $0.024 per credit, 2.4 times the developer rate.
Now convert that into clips. Runway's developer docs price Gen-4.5 at 12 credits per second, so a ten-second clip is 120 credits, or $1.20 bought through the developer portal. Gen-4 Turbo is cheaper at 5 credits per second, which puts the same ten seconds at $0.50.
Then the ground moves. That developer rate is the only per-second price Runway publishes anywhere. What it charges against subscription and free credits inside the app is not on the pricing page, and independent pricing breakdowns from July 2026 put that at 25 credits per second for Gen-4.5. If they are right, that ten-second clip is 250 credits instead of 120, and every clip count you build from the published number gets cut in half. At 12 credits per second a 625-credit Standard month is five ten-second clips; at 25 it is two and a half.
The free plan's 125 one-time credits never renew either. What Runway's pricing page does not do anywhere is convert credits into clips. It sells you 625 of something and lets you discover the exchange rate by spending it.
Sora 2 is the only one of these three with an expiry date, and it is close. OpenAI shut the Sora consumer app on 26 April 2026 and set the API to sunset on 24 September 2026, which was 56 days out from the day I checked these prices.
The other two are products you can build a habit around; this one is a service with a countdown. At $0.10 per second for 720p, or $0.05 on batch, Sora 2 is cheap by the standards of this table, and its dialogue, ambience and sound effects come out of the same generation as the picture. It is also going away before most people would finish learning to prompt it well.
If you want the quality picture rather than the price picture, our Sora 2 test review maps where its physics breaks, using clips other reviewers ran before the app closed. Short version: those reviewers rate it strong on faces and mood, and unreliable on anything with a spinning body in frame.
Runway is the only one of the three that puts commercial use in writing on every plan, free included. Section 4.4 of its terms says "The Company does not claim ownership of any of your Inputs or Outputs" and, in the next breath, "Subject to your compliance with the Agreement, the Company does not restrict your commercial use of your Outputs."
Google is nearly as clear but with a catch worth reading twice. The Gemini API terms say "Google won't claim ownership over that content", then add: "You acknowledge that Google may generate the same or similar content for others and that we reserve all rights to do so." Google's Generative AI additional terms do not address output ownership at all.
Two more things that bite. Runway uses inputs and outputs to train its models by default, and Google's free API tier states that "Google uses the content you submit to the Services and any generated responses to provide, improve, and develop Google products and services and machine learning technologies". Every Veo clip also carries a SynthID watermark on every plan with no opt-out, while Runway's watermark disappears the moment you pay.
For most people it is Veo 3.1 Lite: $0.40 for eight seconds with sound, the cheapest of the six clips in the table. After that, pick by the constraint that actually binds you.
If you need exactly ten seconds in one generation, Runway Gen-4.5 is your only option. If you already pay $19.99 for Google AI Pro, you are holding ten Veo 3.1 Quality clips a month. If a client is paying for the clip, Runway has the cleanest written license.
And if you are considering Sora 2 for anything you still need in October, don't.
My honest take: the interesting difference between these tools in 2026 is not the pixels, it is the pricing logic. One charges per second, one charges per generation, one charges in a currency it never converts for you. Whoever explains that first is doing you a bigger favor than another side-by-side of a robot walking through Tokyo.
Make ChatGPT write like a human with 5 editing rules, not prompt tricks. Raw draft: 19 banned words per 994. Our 35 published posts: 16, all quotes.
NotebookLM Data Tables promises to turn messy data into clean tables. I tried it with a real messy spreadsheet. The catch is at the input, not the AI.
I gave five AI writing tools one brief: write my bio. Each broke differently. Where ChatGPT, Jasper, Copy.ai, Grammarly and Sudowrite fall short.