[Legacy] H3 Continuum Save State

H3 Continuum Save State is the persistence node for the H3_CONTINUUM_STATE object - the raw audio/video latent state that represents "where the video is up to" in a continuation run. It atomically writes that state to disk as a safetensors file plus a JSON sidecar, keyed by a prefix and slot so you can keep several states around.

Inputs: state, prefix (default h3_continuum), slot (1–9999). Outputs: state (passed through) and path (STRING pointing at the written file). It's flagged as an output node, so it can end a graph on its own.

The flow it's designed for: run a H3 Continuum Sampler V2 (or take last_state from a Result node / Session Info), Save State it to disk, close ComfyUI, come back later, Load State the same slot, and feed it as initial_state into a sampler or V1 Join to continue exactly where the last run left off. Because state is the raw latent context - not decoded frames - a saved state continues the chain without any decode/re-encode drift.

Install

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

or via ComfyUI Manager ("H3 Continuum"), then restart. No extra pip packages; needs a MiniMax H3 ComfyUI environment (weights not bundled) and ComfyUI v0.32.0+.

Where this sits in the modern picture: the V3.2.4+ samplers with run_storage = Save + Auto Resume persist per-chunk latents and resume them automatically, which is strictly more convenient for long runs - no manual Save State/Load State wiring. Save State remains the explicit, portable option (it's just a file, you can copy it, share it, or archive it), and it's the mechanism behind the pack's older continuation tutorials. The JSON sidecar is worth knowing about if you ever debug: it holds the state metadata, so a corrupted or mismatched safetensors file becomes visible as a JSON that doesn't match.