H3 Continuum Latent Resize V3.5

H3 Continuum Latent Resize V3.5 is the rare node whose README tells you not to use it for the thing its name suggests. It's a pure utility: it spatially resizes every Continuum physical video LATENT with a ComfyUI Core interpolation method. No decoding, no sampling, no added detail - just geometry. And the pack's own docs are blunt that direct high-ratio interpolation of the 24-channel H3 latent produced persistent artifacts in testing. So why does it exist?

Because latent-space resizing is still a legitimate step in some pipelines, just not the Main path. The README positions it as an advanced utility: if you're experimenting with external latent processors, or you need to match latent geometry before feeding a Second Pass or the Conditioning Bridge, this is the tool that does it while preserving the things that matter - B/C/T dimensions, physical-group order, and LATENT metadata. It's also list-aware, so it resizes every physical group in one shot rather than making you unroll the sampler's list output.

The honest framing: for a 2x upscale with real detail, use H3 Continuum Hi-Res Fix V3.5 (the pixel/VAE round trip) or an external H3 latent processor like LBH. Use this node when you know exactly why you want a latent resize and you accept that you're changing geometry without adding information.

Inputs and output

Only three inputs, all required: video_latents (from the sampler), upscale_method (nearest-exact, bilinear, area, bicubic, bislerp - note Lanczos isn't offered here, unlike the Hi-Res Fix node), and scale_by (default 2.0, range 0.01–8.0).

Single output: resized_video_latents, a list you can feed into H3 Continuum Second Pass V3.5 or the Conditioning Bridge. It doesn't touch audio - there's no audio input at all - so keep your audio latents flowing around it.

Installing it

Same pack as the rest of H3 Continuum - install once, get all six nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum.git

Or search "H3 Continuum" in ComfyUI Manager, then restart ComfyUI. No extra pip dependencies.

The gotcha that matters

Small multipliers are generally fine; it's the aggressive 2x+ interpolation at large square targets that generated persistent artifacts in the pack's own testing. If you try it and see weird plate-like or line artifacts in the output, that's the known failure mode - don't fight it, switch to the pixel/VAE Main path. And remember a latent resize can't add detail it was never given. Interpolation is the cheap rung of upscaling; it makes pixels bigger, it doesn't invent pores or fabric weave.