MiniMax H3 · Video Extend

Every video model gives you a few seconds and stops. Video Extend is the answer to "and then what?" - you feed it the clip you already have, describe the next beat, and it renders more footage that starts exactly where the last one ended, then hands you the whole thing as one file. No second model, no manual stitching in a timeline, no hunting for the frame you cut on.

What it is and why this node

This is the base MiniMax H3 continuation node in the PopNow pack: cloud-only, no weights on your disk, priced from about 5.55 credits ($0.0555) per second of new footage. The source clip's seconds are not billed, which matters - you're paying for the extension, not for re-reading what you already own.

Why not just generate a longer clip in the first place? Because you usually don't know you need longer until you've seen the first pass. Extend is a revision tool. You generate the opening, decide the shot deserves another six seconds, and pay for six seconds.

How it works

The node re-encodes your clip to MP4, uploads it, and asks PopNow for a new segment seeded on the source's last frame. The response is the original clip with the new segment joined on - one file, one VIDEO output. H3 is MiniMax's omni-modal model: picture and stereo audio come out of the same generation pass, so the extension carries sound too, not just motion. That's the reason to reach for this over stitching a silent Wan or LTX continuation with a separate foley pass; the ambience matches because it was generated with the frames.

Uploads are cached by content hash for 12 hours, so tweaking the prompt and re-queuing doesn't re-upload the same clip.

Inputs worth setting

• video_url - the clip to continue. MP4, MOV or WebM, up to 60 seconds. This is a socket, so wire a video loader into it.
• prompt - what happens next: action, camera move, and sound. The tooltip says it plainly, and it's not boilerplate - H3 listens for the audio half.
• resolution - H3 offers 768p and 2k here, defaulting to 768p. Match your source. If you shot the first segment at 768p, extending at 2K gives you a sharp continuation of a soft clip and a visible seam where the sharpness steps up.
• duration_seconds - 4 to 15 on this node. Note the floor is 4 here, where the LoRA variants of H3 start at 3; if you need a 3-second continuation, this is the wrong node.
• last_image_url (optional) - a target end frame, if you want the extension to arrive somewhere specific instead of wherever the model takes it.

Outputs are video (into Save Video) and task id (for the API console or a support ticket). There's no seed widget on this one, so don't count on reproducing a take exactly - capture the result you like and extend from it.

Install

Manager → search PopNow → Install. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Restart ComfyUI, then paste a key from popnow.ai/api/console/keys into Settings > PopNow and restart again so the node list is built from your account. Zero Python dependencies beyond what ComfyUI ships - pyproject.toml has an empty dependency list.

Traps

The 2k tier is where bills jump; the node's price badge updates as you change resolution, so look at it before you queue six of them. The 90 MiB upload limit on video bites on long 1080p sources - trim to the tail you actually need. PopNow 40004 means the model can't serve that combination of settings and names the one to change; PopNow 40201 means credits, not settings.

Last thing: extending a clip that was itself an extension works, but drift compounds. Colour, grain and motion tend to soften with each pass. Two hops and a cut is the practical limit before it starts looking like a copy of a copy.