Both start with a text box. You type a sentence, you wait, a clip comes back. That similarity is why editors keep asking whether Runway replaces a motion graphics tool, or whether a motion graphics tool is a worse Runway.
They are doing opposite jobs. A video generator invents pixels: a shot that never existed, a drone pass over a city, a close-up of coffee hitting a cup. Butter Motion builds a composition: shapes, text, numbers and timing that were specified, then rendered. The moment your graphic has to read 47.3% and spell a surname correctly, that difference stops being philosophical.
Everything below about Runway, Pika, Luma and Kling comes from their own pages, read on 2 September 2026. Prices move, so read their cards before you pay. We have not benchmarked their output quality, because we do not run them on deadlines and a guess dressed as a test helps nobody.
A generative video model works in frames. It has no concept of a text layer, a fill colour or a data value; it has a probability distribution over pixels that, if all goes well, look like the words you asked for. This is why the class is extraordinary at things with no correct answer, a dragon over a lake, a fog-lit alley, a product shot from an angle you never filmed, and why it stays awkward at things where one answer is correct and the rest are wrong.
Butter Motion goes the other way. Your prompt is planned into a composition, that composition renders on your machine, and it is still there afterwards. That is what makes the customization controls possible: unlock them on the result screen and you get the real properties, colours, text lines, numbers, toggles, fonts, with a live preview as you move them. Nothing is re-generated. The approved composition is rewritten from your values.
There is a plainer way to say this. In a generator, the text in your clip is a picture of text. Here, it is text.
Generators price by output: seconds, resolution, model. That is honest, and it is also the thing that shapes how you work with them.
Runway's Standard plan is $15 a month, or $12 when billed annually, and carries 625 credits a month. Their own page does the arithmetic for you: "625 credits monthly = 52s of Gen-4.5". Gen-4.5 runs 60 credits per 5 seconds, and their FAQ puts it at 12 credits per second of generated video. Pro is $35 (or $28 annually) for 2 250 credits, Max is $95 (or $76 annually) for 9 500. On Standard and Pro the credits do not roll over; Max carries unused credits one month forward and is also where "Studio-grade formats (ProRes, image sequences)" appear.
Pika sells 80 monthly credits free, 700 on Standard at $8 a month billed yearly, 2 300 on Pro at $28, 6 000 on Fancy at $76. Their table is granular: a 1080p clip costs 40 credits at 5 seconds and 80 at 10, a 720p clip 20 and 40. The free tier is 480p only and image-to-video only. Durations for text-to-video are 5 or 10 seconds, with Pikaframes stretching to 20-25 seconds.
Luma's Plus plan is $30 a month or $300 a year for 10 000 credits, Pro $90, Ultra $300. Ray3.2 at 1080p costs 400 credits per 5 seconds. Running Kling 3.0 through Luma is 40 credits a second at 1080p and 147 at 4K, so the length and the resolution are the bill.
Butter Motion charges for the thinking, not the output. The tier is the whole price: 7 credits on Flash, 10 on Balance, 15 on Ultra. Length, aspect ratio, resolution, frame rate and transparency do not move it, and 3D on Ultra costs nothing extra. A 3-second 720p clip and a 20-second 4K clip cost the same. The rules are in How credits work and Quality tiers and 3D.
| Butter Motion | Runway | Pika | Luma | |
|---|---|---|---|---|
| What you pay for | Quality tier only | Seconds, model, resolution | Seconds, model, resolution | Seconds, model, resolution |
| A 20s 4K clip vs a 3s 720p one | Same price | Longer and larger costs more | Longer and larger costs more | Longer and larger costs more |
| Clip length | 3 to 20 seconds | Priced per second | 5 or 10s, Pikaframes to 20-25s | Priced per second |
| Alpha channel | MOV, ProRes 4444, every plan | Not stated on the pricing page | Not stated on the pricing page | Not stated on the pricing page |
| Free tier | None | 125 one-time credits | 80 credits a month | Not listed on the pricing page |
A lower third, a callout, a corner logo and a chapter label share one requirement: they sit on top of footage. That needs a real alpha channel.
Butter Motion writes a transparent generation as MOV in ProRes 4444, on every plan and every tier, with no surcharge, and a clip with a background comes back as MP4. You choose with one chip before you generate. See Background and transparency. On the timeline, a transparent clip is placed on a track above your footage so it keys over what is underneath, which is covered in Rendering to the timeline.
On the pricing pages of Runway, Pika and Luma we found no mention of transparent output or an alpha channel. That is not a claim that none of them can do it anywhere. It is a note that if alpha is what your edit needs, nobody is selling it to you on the page where you pick a plan, so confirm it before you build a workflow on one.
Nothing lands right the first time. What matters is what the second pass costs, and this is where the two models diverge hardest.
With a generator, a small change is another generation. The name is misspelled, so you regenerate, and seconds times resolution bills again. Worse, you get a different clip. Generation is stochastic, so take two is not take one with a corrected surname; the light moves, the framing shifts, and now the graphic at 00:14 does not match the one at 02:31 even though they were meant to be the same lower third with a different name.
In Butter Motion the second pass is usually free. The first tweak on every generation costs nothing while your subscription is active, and it edits the existing clip rather than rebuilding it. The customization controls are free for subscribers every time. Re-rendering and exporting never cost credits, so the same result can go out at three resolutions and still be one generation. A failed generation refunds itself automatically.
Consistency is the quiet part. Eleven lower thirds in one cut have to match each other, and matching comes from a composition you copy and adjust, not from eleven rolls of dice.
We are not the tool for a large part of what people want video AI for.
Kling is worth a note. Its models turn up inside other products, priced in Luma's tables and listed on Runway's Standard plan, so pick a generator on its output rather than on the app around it.
If your deliverable is a shot, use a generator. If your deliverable is a graphic that carries information, a chart, a counter, a title, a lower third, a price card, use something that keeps the information editable and puts it on your timeline with alpha.
Most editors need both, and they are cheap enough to sit side by side. A generator for the plate, a plugin for the layer on top of it. Our side is two tiers, Pro and Studio, with 500 and 2 000 credits a month, laid out in Pro and Studio.
Neither tool replaces a motion designer. The routine layer is what changes hands: name plates, chapter cards, counters and charts an editor can close alone. The title sequence and the brand system still go to a designer.
The test that settles it is small. Ask for a lower third with a surname you have to spell right, then ask for the same one with a different name. If the two match, you are in the right tool.