Is Mischief 1 based on Wan 2.2?

Yes. TBC built it on Alibaba’s open Wan 2.2 model, adding a small post-training layer (under 0.1% more parameters, per Unite.AI).

Is it better than Wan 2.2?

TBC claims the same quality at about five times the speed, but no independent comparison has been published. Wan 2.2 is more flexible (text-to-video, open weights); the model is much cheaper per second on fal.

How much cheaper is Mischief 1 than Wan 2.2?

On fal, the model costs $0.005/s (480p) and $0.01/s (720p), versus $0.04/s and $0.08/s for Wan 2.2 A14B image-to-video – one eighth of the price.

Can I run it locally like Wan 2.2?

None of the sources we checked mention downloadable weights for it; it is offered as a hosted model on fal, OpenRouter and sites like this one. Wan 2.2 weights are open under Apache 2.0.

Mischief 1 vs Wan 2.2: An Honest Comparison

Last updated: 2026-11-04

ClipWhim is an independent product and is not affiliated with, endorsed by or sponsored by The Biological Computing Co. (TBC). “Mischief 1” is used only to describe the third-party model ClipWhim runs, accessed via API; the name belongs to its owner.

Mischief 1 vs Wan 2.2 is not a comparison between two unrelated models: Mischief 1 is built on top of Wan 2.2. The Biological Computing Co. (TBC) took Alibaba's open Wan 2.2 video model and added a small post-training layer that it says was derived from neuron research (The Next Web, OrigoTech). So the real questions are: what did TBC change, what did it give up, and when should you use which? This page compares them honestly, using only published specs. You can try the model in our online generator.

Side-by-side at a glance

	Mischief 1 (TBC)	Wan 2.2 (Alibaba)
Released	October 30, 2026	July 2025
Relationship	Built on Wan 2.2 + TBC post-training layer	Base model
Modes	Image-to-video only	Text-to-video, image-to-video, text+image-to-video (TI2V-5B), speech-to-video (S2V-14B)
Resolution	480p (832×480) or 720p (1280×720)	480p and 720p (A14B models); 720p at 24 fps (TI2V-5B)
Aspect ratio	16:9 only	Depends on the endpoint / settings
Length	3 or 5 seconds	Configurable frame count (81 frames by default on fal)
Frame rate	16 fps	Configurable on fal; 24 fps for TI2V-5B
Audio	None	None in T2V/I2V; S2V is driven by speech input
Open weights	None published, as far as our sources show (hosted API)	Yes, Apache 2.0
Run locally	Not that we know of	Yes (TI2V-5B runs on a consumer GPU like an RTX 4090, per Alibaba)
fal price, 480p	$0.005 / second	$0.04 / second (A14B image-to-video)
fal price, 720p	$0.01 / second	$0.08 / second (A14B image-to-video)
Speed	TBC claims ~5× faster than Wan 2.2 – not independently verified	Baseline

Sources: fal API, OpenRouter, Wan 2.2 on GitHub, fal Wan 2.2 A14B image-to-video, checked November 4, 2026. Prices can change.

What TBC changed

According to Unite.AI as summarized by OrigoTech, the model is a post-training layer around Wan 2.2's diffusion transformer blocks that adds under 0.1% to the parameter count – it modifies Wan 2.2 rather than replacing it. TBC says the result runs about five times faster at the same quality.

TBC has not disclosed which Wan 2.2 variant, hardware, resolution or step count its speed figure is based on, and no independent benchmark had been published at the time of writing (TNW, OrigoTech). OrigoTech also notes that no direct, controlled head-to-head comparison has been published. We have not run one either, so this page makes no quality claim in either direction.

Price: the clearest difference

Price is the one difference that is fully public. On fal, the model costs $0.005/s at 480p and $0.01/s at 720p; fal's Wan 2.2 A14B image-to-video endpoint costs $0.04/s at 480p and $0.08/s at 720p. That makes a 5-second 720p clip $0.05 vs $0.40 on fal – one eighth of the price. See TBC model pricing for a full breakdown.

One caveat: Wan 2.2 is open-weight under Apache 2.0, so if you have your own GPUs you can run it without paying a per-second API fee at all – you pay for compute and setup time instead.

Flexibility: Wan 2.2 does more

Text-to-video: Wan 2.2 has it; the model needs a first-frame image.
Local and custom workflows: Wan 2.2 weights can be downloaded, fine-tuned and used in tools like ComfyUI; the model is available only as a hosted model.
Output shape: The model is fixed to 16:9 and 3 or 5 seconds at 16 fps.

Which should you use?

Choose Mischief 1 if…

You already have a still image and want a short landscape clip.
Low cost per try matters – drafts, moodboards, many variations.
You want no setup: use the web generator or a simple API call.

Choose Wan 2.2 if…

You need text-to-video or vertical / custom aspect ratios.
You want to run the model locally, fine-tune it, or build a ComfyUI pipeline.
You need settings the model does not expose, such as frame count or frame rate.

Because the model starts from Wan 2.2, prompting habits carry over. The Wan 2.2 guide on fal recommends "motion description + camera movement" for image-to-video prompts – the same approach in our prompt guide.

Looking beyond Wan? See alternative models, or learn more about the company in TBC Mischief explained.

Frequently asked questions

Is Mischief 1 based on Wan 2.2?: Yes. TBC built it on Alibaba’s open Wan 2.2 model, adding a small post-training layer (under 0.1% more parameters, per Unite.AI).
Is it better than Wan 2.2?: TBC claims the same quality at about five times the speed, but no independent comparison has been published. Wan 2.2 is more flexible (text-to-video, open weights); the model is much cheaper per second on fal.
How much cheaper is Mischief 1 than Wan 2.2?: On fal, the model costs $0.005/s (480p) and $0.01/s (720p), versus $0.04/s and $0.08/s for Wan 2.2 A14B image-to-video – one eighth of the price.
Can I run it locally like Wan 2.2?: None of the sources we checked mention downloadable weights for it; it is offered as a hosted model on fal, OpenRouter and sites like this one. Wan 2.2 weights are open under Apache 2.0.

Try the AI video generator

Upload an image, write a motion prompt and get a 3–5 second widescreen clip. Failed generations are refunded automatically.

Generate a video