Nodes/ComfyUI-MaxedOut/LTX Empty Latent Video MXD
ComfyUI Node

LTX Empty Latent Video MXD

The LTX empty latent that doesn't make you do the 8n+1 math

By Maxed-Out-99·Created about a year ago·Updated 21 days ago· 9
LTX Empty Latent Video MXD
    • latent
    • length
    length97
    batch_size1
    width640
    height384

    LTX Video is the speed king of local generation - seconds per clip where WAN takes minutes - but its latent format has rules that punish you for being sloppy: widths and heights must be multiples of 32, and frame counts must land on 8n+1 (9, 17, 25 … 97, 121). Type 100 frames into a plain empty-latent node and you'll either error out or quietly get a broken video. LTX Empty Latent Video MXD is the pack's answer: it takes your width, height, and frame count and builds a correct LTX latent, snapping your values to the right alignment automatically.

    What it does

    Required inputs:

    • length - target frame count (default 97). The tooltip is the whole game: "Must be 8n+1 (e.g. 25, 49, 73, 97, 121, 201)." The node rounds your input to the nearest valid 8n+1, so you don't have to.
    • batch_size - how many latent videos to make (default 1).

    Optional:

    • width (default 640) and height (default 384) - and note the tooltips: these are stage 1 dimensions. In LTX's two-stage I2V workflow you generate a small latent first, then upscale and refine, so the author wired this node to connect straight from the LTX Image Scaler's stage1_width / stage1_height outputs. That's the intended pattern for I2V - follow it and you don't have to remember stage-1 vs stage-2 sizes at all.

    Outputs are latent (a proper LTX latent: 128 channels, 32× spatial compression, 8× temporal compression) and length (the actual, snapped frame count - handy if you want the real number downstream instead of the number you typed).

    Where it fits

    Drop it in front of an LTX sampler (the pack ships LTX sampler nodes too, and this one feeds them directly). For text-to-video it's a straight empty-latent replacement. For I2V, connect the scaler's stage-1 outputs and let the node handle alignment. It's a "make the fiddly thing just work" node in the truest sense - LTX's constraints are exactly the kind of detail that eats an afternoon when you're new to the model.

    Gotchas

    • Don't fight the snapping. If you need exactly 101 frames for some reason, this node will round it to 97 or 105; LTX genuinely can't do arbitrary lengths, so the snap is a feature.
    • The defaults (640×384, 97 frames) are LTX-friendly, not quality-optimal - they're a solid low-memory starting point, and the KB's guidance is that LTX rewards iterating at low res and refining rather than brute-forcing big latents.

    Install

    ComfyUI Manager → search Maxed Out → install → restart, or:

    cd ComfyUI/custom_nodes
    git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut
    

    No requirements.txt, no models to download. Under MXD/Latent in the menu. If LTX has been giving you "invalid dimensions" headaches, this is the node that quietly removes that whole class of error.

    CategoryMXD/Latent

    Inputs (4)

    NameTypeDefaultDescription
    lengthINT979–1025Number of frames. Must be 8n+1 (e.g. 25, 49, 73, 97, 121, 201).
    batch_sizeINT11–4096Number of latent videos in the batch.
    widthoptINT64032–8192Stage 1 width. Connect the LTX Image Scaler stage1_width output for I2V workflows.
    heightoptINT38432–8192Stage 1 height. Connect the LTX Image Scaler stage1_height output for I2V workflows.

    Outputs (2)

    NameTypeDescription
    latentLATENT
    lengthINT