H3 Auto Chain Load Latent

H3 Auto Chain Load Latent is the pack's memory: it loads the previous clip's saved H3 latent so the next clip can build on it. It looks like a tiny node - one LATENT output - and it is, but it's also doing one clever thing that makes the whole auto-chain design work.

The clever thing: it knows when there isn't a previous clip. On clip 1, or any time the previous latent is missing, it quietly returns no latent at all, and the Motion Context node sees that and passes its conditioning straight through with zero context trim. So the same workflow that starts a chain also works for the first clip, no special-casing on your part. It only loads a real latent when a real previous clip exists.

The inputs

Three required, one optional, and honestly only the optional one matters for an auto-chain:

• chain_config (optional) - the H3_CHAIN output from H3 Auto Chain Audio. Connect it and the node picks the correct previous clip slot automatically: it loads clip N-1 when you're rendering clip N. This is the field that turns the node from manual into automatic.
• latent_path - the folder the latents live in, default h3_context. You shouldn't need to touch it with chain_config connected.
• clip_index - a manual clip number to load, used when chain_config isn't wired. With chain_config connected it's overridden.
• reset - when true, the node returns nothing regardless. It's the "start fresh" switch; the auto-chain flips it on run one.

Where the output goes

There is exactly one correct destination for the LATENT output: the context_latent input on MiniMaxH3AutoChainMotionContext (or the original H3 Motion Context node if that's what your workflow uses). Do not connect it to a VAE decode, do not feed it to a sampler, do not get creative. The latent is H3's paired video/audio representation, and the only thing that understands it is the Motion Context node. This is the single most common way people break this pack, and it's the kind of error that shows up as garbage pixels rather than a clean crash.

Troubleshooting

• "The addon cannot find a previous latent" - check that chain_id is unchanged, start_clip is correct, the previous clip actually saved, and that this addon's Load Latent node (not the stock ComfyUI Load Latent node) is the one wired into Motion Context.
• Clip renders fine but is visually disconnected - that's the missing-latent fallback working as designed: the chain still runs, Motion Context just gets bypassed for that clip. Render the missing clip with the same chain_id and re-stitch.
• Continuity keeps breaking at clip 2 specifically - check the previous clip's latent file actually exists under output/h3_context/. Missing previous videos don't stop a partial render, but missing latents always cost you the continuity.

Install is the pack install - Manager, search "H3 Motion Context", or git clone https://github.com/Ltamann/ComfyUI-H3-Motion-Context-Auto-Chain-addon into ComfyUI/custom_nodes/, then restart. No extra dependencies. And yes, it needs safetensors, which your H3 workflow already ships with.