Gigachad LTX Resolution Picker

Gigachad LTX Resolution Picker is the same idea as the regular Resolution Picker, but built for LTX Video - and the difference is not cosmetic. LTX's spatiotemporal VAE compresses video differently from SD's image VAE: 32x in space and 8x across time, into 128 latent channels. If you fed an LTX video model an SD-shaped empty latent, nothing would line up. This node constructs the latent with LTX's exact shape so you don't have to know any of that math.

The inputs

• width (default 1280, step 32) and height (default 720, step 32) - the frame dimensions. LTX natively renders at half the requested resolution and upscales internally, which is exactly why 1280×720 is the community default rather than 1920×1080: you get 720p-class output without burning VRAM on a 1080p latent.
• length (default 97) - the frame count. The step is 8 and the minimum is 9, which tracks LTX's temporal compression: the latent's time axis is ((length - 1) // 8) + 1, so a 97-frame clip becomes a 12-slice latent. The odd-looking default is deliberate - LTX wants lengths that survive division by 8 plus 1.
• batch_size (default 1) - number of video latents to create.

The outputs

Exactly three, same names as the image picker: WIDTH (INT), HEIGHT (INT), and LATENT (a torch.zeros tensor shaped [batch, 128, (length-1)/8+1, height/32, width/32]). Wire the latent into your LTX sampler's latent_image and the ints wherever you need them - which, for video workflows, is often a frame-count or duration calculation downstream.

How it fits the LTX picture

If you're new to LTX, the KB's model panel sums up the family: Lightricks' speed-first video line, where the whole selling point is seconds-long drafts on consumer VRAM. Getting the latent shape right is the single most common setup mistake in video pipelines - a latent with the wrong channel count or temporal axis is what produces instant "shape mismatch" errors. This node removes that class of bug entirely, and the defaults (1280×720, 97 frames) are sensible starting points for the current LTX releases.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/comfyui-gigachad.git

or install comfyui-gigachad via ComfyUI Manager, then restart. No extra dependencies.

Honest note: you could hand-roll this latent with EmptyLatent math nodes, but you'd be doing the (length-1)//8+1 and //32 arithmetic yourself and getting it wrong once in a while. For LTX-specific work this picker is the better tool - it's a genuinely useful node, just for a narrow audience. If you generate images, not video, you want the regular Gigachad Resolution Picker instead.