Install the kit

One folder with everything inside: its own Python, ComfyUI, the models, the workflows and the tools. Nothing to install.

1. Download the kit

models_in_motion.zip · 68 GB, Dropbox

It is big: start the download early. Unzip it on a fast SSD with about 70 GB free (NTFS or exFAT, not FAT32). You get a folder called models_in_motion.

2. Update the tools

models_in_motion_tools_2026-10-30.zip · 90 KB

In your models_in_motion folder, rename the old tools folder to tools_old.
Double-click the zip to open it, and drag the tools folder inside it into models_in_motion.

This only touches the tools. ComfyUI, the workflows and your renders stay as they are.

3. Start

`CHECK.bat`	Before the first time: shows the GPU and driver, and whether PyTorch sees the GPU. If it says NO GPU, update the NVIDIA driver (580 or newer).
`START.bat`	Starts ComfyUI. The browser opens on `http://127.0.0.1:8188`. Close ComfyUI by closing its black window.
`START_safe.bat`	The same, without SageAttention. Use it if START fails on "triton" or "sageattention".

The first render after starting is slow (the models load from disk); the next ones are fast.

The workflows

In ComfyUI: left bar → Workflows.

01	LTX 2.5, text to video	a sentence becomes a clip, with sound
02	LTX 2.5, image to video	a frame becomes a clip
03	LTX 2.5, first and last frame	two frames; LTX finds the movement between them
04	Qwen 2.1, text to image	a sentence becomes a frame
05	Qwen 2.1, image edit	change a frame with words
06	Qwen 2.1, background removal	a frame with a transparent background
07	LTX 2.5, grey box to render	a reference video drives the movement; the prompt and a frame set the look
08	footage to movement guide	your clip becomes depth and a skeleton: only the movement survives
09	example: footage depth drives the render	07 fed with 08's depth

Many takes of one shot

In the small box next to the blue Run button, type 13 and press Run once. ComfyUI queues 13 renders, each with a new seed: 13 takes of the same shot. They appear one by one in models_in_motion\ComfyUI\output\video.

If they come out identical: inside the big node, find RandomNoise and set control after generate to randomize.

Where things are

`ComfyUI\output\video`	your videos
`ComfyUI\output`	your images
`tools`	RUN, INDEX, the scripts (see Tools)
`examples`	takes to practise on: a subway shot, a wave

Every image and video carries its workflow inside: drag it back onto ComfyUI to reopen exactly how it was made.