FeiHou Easy H3 Output

The main Easy H3 node hands you a MINIMAX_H3_CONTEXT - one opaque wire that carries conditioning, a latent, two VAEs, an FPS value and the final prompt. That's great until you want to actually sample and render, because ComfyUI's sampler nodes don't speak H3 Context. FeiHouEasyH3Output unpacks the context back into the standard ComfyUI types so you can plug into the exact KSampler → VAE decode → video-save chain you already know. It's the node that turns the pack from a one-box toy into something you can bend.

How it works

Nothing clever, and that's the point. Give it h3_context and it returns the six pieces the main node stashed inside:

• positive - CONDITIONING, straight into the KSampler's positive pin.
• latent - LATENT, the empty (or keyframe-seeded) latent at your chosen resolution and frame count. Into the sampler's latent.
• video_vae and audio_vae - the two VAE objects. Video VAE decodes frames; audio VAE decodes the soundtrack. H3 generates audio jointly with video, so both come back to you here.
• fps - a FLOAT, exactly the FPS you set on the main node (default 24), so you don't have to re-enter it when saving the video.
• prompt_preview - a STRING with the final prompt after any optimizer expansion or reference inference, ready to feed the pack's Prompt Preview node.

It's a plain unpacker: it validates the input is a real MiniMaxH3Context, then returns the fields. There are no settings and no model weights. If you've seen this pattern before it's the same idea as any "extract" node in a pack that uses custom pipe types - the pipe keeps the graph tidy on one side and the extractor makes the standard side work again.

Wiring it up

The canonical downstream chain is: FeiHouEasyH3Output → KSampler (positive, latent) → the video VAE's decode → a save node like VideoHelperSuite's VideoCombine (feeding it the fps), and the audio VAE + latent if you want the soundtrack out separately. The two-pass crowd uses it per pass. If you're doing straight image mode, the same outputs work - you just decode one frame.

How to install it

Pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/FX-FeiHou/ComfyUI-FeiHou-Easy-H3

restart, or ComfyUI Manager → search "ComfyUI-FeiHou-Easy-H3". No pip dependencies, no models of its own - the models are H3's, and the H3 Community License territory restrictions (no US, EU, UK, Korea) apply to the whole stack above it.

Common issues

The failure mode is connecting the wrong context: feed it a bundle instead of a context and it throws, so make sure the wire comes from the main node's h3_context output, not the Loader's h3_bundle. And a heads-up for the audio path - H3's audio VAE output isn't something every save node knows what to do with. If your video saves silent, the soundtrack is decoded by the audio VAE separately; that's expected behavior, not a lost render. On the VRAM front, the main node's force_offload does its cache cleanup right before VAE decode, so this is the stage where that option earns its keep.