MatlowAI adds MiniMax H3 video controls through files of around 14 KB

You have generated an AI clip that fits the brief, but the movement is too sluggish. You start rewriting the prompt—and the composition and main character change along with the speed. MatlowAI proposes another approach: create a control from two descriptions of movement, then adjust its strength with a slider. For developers of video services, the appeal is the ability to turn a recurring edit into a dedicated interface element.

The project is called Dials and works with MiniMax H3 through ComfyUI. Its documentation labels it as alpha. That makes it worth trying on familiar scenes, but it is too early to promise clients precise direction from a single slider.

How two descriptions become a control

To create a Dial, you supply a pair of prompts that differ in one property. For example, the character stands still in the first and runs in the second. The tool turns the difference between their internal text representations into a file of around 14 KB. Negative strength shifts generation toward the first description; positive strength shifts it toward the second. No separate model training is required. This is how the mechanism is described in the Dials documentation.

A hypothetical assignment makes the practical value easier to understand. Suppose a team is producing several versions of an advertising scene featuring the same character. The client keeps asking for more movement. Instead of trying another set of phrasings, the team could try the same Dial at different strengths. If the effect proves sufficiently consistent, it makes sense to save that control alongside the project settings.

A small file, however, does not mean a small video model or inexpensive generation. It specifies a change in the behavior of a system that is already loaded. Estimating costs requires accounting for the entire process of producing a clip: hardware, rendering time and the number of failed attempts. The size of a Dial alone tells you nothing about those expenses.

What you can build in ComfyUI

The ComfyUI-MAINodes package includes a Dial Rack panel for multiple controls. You can experiment with combinations of movement, mood and visual styling. The repository also contains ready-made workflow examples, including one for Pika H3 Flex. The authors describe Dials as controls that require no fine-tuning and explicitly note the tool’s experimental status. The project’s main README confirms this.

Strength curves over the duration of a clip are also supported. Here, it is important to distinguish between being able to draw a curve and being able to achieve exactly the intended visual effect. A slider with a timeline does not, by itself, demonstrate that a brief stylized effect will land precisely on a beat or preserve every detail in adjacent frames.

To test it, I would choose a simple scene with a clear action. I would first generate a baseline, then compare several values of a single control while keeping all other settings unchanged. Next, I would repeat the experiment with different random seeds. The questions are specific: does the intended property change, are the character and composition preserved, and how many versions have to be discarded?

The limits of a convenient interface

The authors explicitly warn about side effects at high strengths: the scene can be restructured and characters can change appearance. The documentation suggests roughly +0.5 to +1 for typical use; problems can emerge around +1.5, and they appear earlier in the short generation workflow for Pika H3 Flex. These are the developers’ guidelines, not a universally safe setting.

The original post’s claim of full compatibility with third-party LoRAs therefore needs to be tested in the specific setup. The technical ability to connect multiple tools does not guarantee that their effects will work well together. For a product, what matters is the final yield of usable clips, rather than the number of available switches.

The practical next step is a small comparative test on your own tasks. Decide in advance what edit the Dial should make and which changes are unacceptable. If the control reduces the number of attempts without losing necessary details, it can become part of the workflow. For now, this is a useful experimental tool with a clear concept and an open implementation; each team will need to measure its time savings and degree of control for itself.