TBC mischief 1: AI Video at 2.5 Cents a Clip Makes Moving Storyboards Cheap Enough to Throw Away
A five-second AI video clip now costs less than the paper you would sketch it on. TBC's new mischief 1 model charges half a cent per second of video at its lower resolution, which puts a whole page of moving storyboard panels under a dollar. (You have to pick the lower setting yourself; the default is the sharper 720p, at twice the price.) Its maker says that price comes from software derived from living neurons grown in a lab.
That second claim is the one people will argue about. The first one is the one that should change how you work.
What mischief 1 is
The Biological Computing Co., a San Francisco company that goes by TBC, launched mischief 1 on October 30. It is an image-to-video model: you give it a still picture and a sentence describing what should happen, and it returns a short clip that starts from your picture.
It is built on Wan 2.2, a widely used video model whose files are free to download, and TBC adds its own speed layer on top. The company says the result "generates video 5x faster than its base model" with the same quality.
You can run it two ways today. On fal, a pay-per-use hosting service for media models, it costs $0.005 per second at 480p and $0.01 per second at 720p. A five-second clip is 2.5 cents at the low setting and 5 cents at the high one. It is also listed on OpenRouter from $0.005 per second.
The shape of what you get back is narrow:
- Three or five seconds long.
- Widescreen only: 832 by 480 or 1280 by 720 pixels.
- 16 frames per second (fps), fewer than the 24 most film work uses, so motion looks slightly steppier.
- No audio is mentioned anywhere in the docs.
Those limits matter, and I will come back to them. First, the price.
Why the price is the story
Most of the cost of AI video has never been the final clip. It is the twelve clips you made before it, the ones where the camera drifted the wrong way, the character turned into someone else, or the motion felt dead.
For a sense of scale, last week's AI motion-capture project anymove used a five-second clip from MiniMax's H3 Max on fal at about 26 cents. That is a strong model, and 26 cents is not much for a shot you keep. It adds up fast for shots you throw away.
At 2.5 cents, mischief 1 changes the math of the throwaway. Ten attempts at one idea cost a quarter. A storyboard page with twelve panels, animated once each at 480p, costs 30 cents.
That is the price of a thumbnail sketch, and I think it should be used like one. Thumbnails exist so you can make the big decisions cheaply: framing, direction, rhythm, whether the idea works at all. Nobody frames a thumbnail. You make forty, keep three, and spend real effort on those.
Video AI has lacked that stage. Every attempt cost enough that you prompted carefully, hoped, and then talked yourself into whatever came back. A sketch-priced model lets you ask the cheaper question first: does this shot even move the way I imagine?
The neuron claim, and why you can ignore it for now
TBC's press release says the company grows living neurons in a lab, measures how they process information, and turns what it learns into software that runs on ordinary AI hardware. That software, the company says, "adds less than 0.1% to the base model's parameter count," meaning it barely changes the size of the model. The neurons stay in the lab.
It is a striking story, and the press release does not describe how the method works in technical detail. The "5x faster" figure comes from TBC, with no independent benchmark attached. The headline calls it the "World's First Publicly Available Neuron-Derived AI Video Model," which is also TBC's own description.
For a working creator, none of that changes the decision. You are not buying a theory. You are buying clips at a published price, and you can judge those clips with your own eyes in an afternoon. Treat the speed claim as unproven and the price as real, because the price is the part printed on the fal page.
Put this into practice: a moving storyboard for under a dollar
Here is how I would use mischief 1 on a real job, say a 30-second spot or a short film sequence. The goal is an animatic: your storyboard cut together in order, with rough motion and sound, so you can feel the timing before anything is shot or rendered properly.
1. Start from frames you already have. Export your storyboard panels, concept frames or rough key art as individual images. The model starts every clip from your image, so the composition you drew is the composition you get. Keep them horizontal, because that is all the model makes. fal crops your image to fit by default; switch the crop setting to "resize" only if you would rather stretch than lose the edges.
2. Run everything at 480p and three seconds first. You are testing motion, not detail. Three-second clips at 480p cost 1.5 cents each. For timing questions (does the push-in feel slow? does the turn land?) three seconds is often enough.
3. Keep your own words. By default, fal rewrites your prompt to fit the image before the model sees it. That is helpful for a quick result and unhelpful when you are comparing attempts, because you no longer know what changed. In the settings, set prompt processing to "original" so your wording goes in as typed. fal's own note says "Long, specific prompts steer WAN best," so describe the motion plainly: who moves, which way, how fast, what the camera does.
4. Change one thing at a time. fal lets you set a seed, a number that fixes the randomness. Keep the seed the same and change only the prompt, or keep the prompt and change only the seed. That turns a pile of clips into an actual comparison.
5. Cut it together before judging any single clip. Drop the clips into your editor in story order with a temp track, a stand-in piece of music or voiceover. A shot that looks weak on its own can work in sequence, and a lovely clip can stall the cut. This is where animatics earn their keep.
6. Promote only the winners. Once you know which shots work and how they should move, take those few into whatever model you trust for finals, with a prompt you have already proven. Finals still cost what finals cost. You just make far fewer of them.
A rough budget for a 30-second spot with fifteen shots and four attempts per shot: sixty three-second clips at 480p is 90 cents. Even doubled for the shots that fight you, that is less than one coffee for a moving animatic you can show a client.
Where it breaks
It is not a finishing tool. Sixteen frames per second reads as slightly choppy next to 24 or 30 fps footage, and 720p tops out below what most delivery specs ask for. You can add in-between frames and enlarge the picture with other tools later, but at that point you are polishing a sketch.
Widescreen only. If your job is vertical social video, mischief 1 cannot frame it. You would be animating a horizontal clip and cropping, which wastes most of the picture.
Five seconds is the ceiling. Longer moves need stitching, and stitched AI clips rarely match at the seam.
Your prompt is screened. fal's documentation says TBC's policy service screens every prompt and image. For most work that will never come up. If your project deals with violence, horror or anything near a content line, expect refusals you cannot see the reasons for.
The terms are thin. fal labels the model for commercial use, but neither fal's page nor TBC's press release spells out terms beyond that label. For animatics and internal boards, that is fine. Before a mischief 1 clip appears in anything you deliver, ask for the terms in writing.
One provider behind the curtain. On OpenRouter, every request goes to a single company running the model, and the share of time it was listed as working was under 88 percent when I checked. That is fine for an afternoon of sketching and risky for a deadline.
Quality claims are TBC's. The "same quality, five times faster" line has no outside test behind it yet. One studio, Wonder Studios, is quoted saying the speed and low cost let it push its work further, which is exactly the use case here, and also exactly what you would expect a launch quote to say.
What to do with it this week
Take one sequence you are already working on, five or six panels, and animate each one three times as three-second clips at 480p. That is under 30 cents. Cut the best version of each into a sequence and watch it with sound.
You will learn two things quickly. Which of your shots wants to move the way you drew it, and which ones were only ever going to work as stills. Both answers are worth far more than the quarter they cost. And once you know a shot works, you can pay full price for it without the usual knot in your stomach.