Midjourney Load Video (Ace Data Cloud)

What it is

The video counterpart to the pack's image loader, and the second half of the Midjourney Videos flow. The videos API returns a URL; this node turns that URL into a native ComfyUI VIDEO you can hand to a save node, an interpolator, or anything else in the graph that accepts a video.

Worth being blunt about the use case: Midjourney's video output is a finished artefact, not a latent. You're not going to re-sample it. What you'd actually want this for is chaining - saving the clip beside your local Wan or LTX renders, running it through a frame-extraction or upscaling step, or building an evaluation grid of several generations side by side. If that's not your plan, the URL and a download are fine.

How it works

You give it media_url, a STRING, and it applies the same downloader the image loader uses: https only, hostname restricted to the Ace CDN domains, no redirects, a 128 MiB cap, three retries on a connection failure. The bytes then go into ComfyUI's own video object (VideoFromFile) and come out the video output as a native VIDEO.

That last detail is the important one, and it's why this node has a hard requirement the rest of the pack doesn't: ComfyUI 0.39.0 or newer. The native video type comes from comfy_api.latest, so on an older install you don't get a graceful error, you get an import failure when the module loads. There's no pip package substitute - the pack deliberately piggybacks on ComfyUI's own media libraries rather than adding PyAV or ffmpeg wrappers of its own.

Also note it does not decode anything at load time. It wraps the bytes; the downstream node that actually reads frames does the work. That keeps the node cheap, and it means a broken container shows up later in the graph, not here.

Inputs and outputs

• media_url - the only input, and yes, it's the same field name as the image loader's. Copying the URL from the wrong node is the most common mistake here.
• video (VIDEO) - the clip. Wire it to Save Video, a frame loader, or an upscaler.

Empty input raises "Task is pending; retrieve the same task ID again." - which is the pack's way of saying your Midjourney Videos node hasn't finished. Go poll it rather than re-running the paid job.

Install

ComfyUI Manager → search Midjourney → publisher acedatacloud, package Midjourney by Ace Data Cloud → install → restart. For a source install:

cd ComfyUI/custom_nodes
git clone https://github.com/AceDataCloud/MidjourneyComfyUI
cd MidjourneyComfyUI
python -m pip install -r requirements.txt

requests and jsonschema, nothing heavier. No model files, no ffmpeg build step. The loader needs no API key - it only talks to the CDN.

Common issues

Import error on ComfyUI startup mentioning comfy_api. You're below 0.39.0. Update ComfyUI; the rest of the pack's tool nodes would probably still work, but this one can't.

The 128 MiB ceiling. It's generous for an image and not generous for video. A long or high-resolution clip that the CDN serves whole can exceed it, and the error is a flat "Media exceeds the 128 MiB limit." If you hit it, take the first_media_url out of ComfyUI and download the file yourself, or extend in shorter segments on the API side.

"The media link does not contain a valid image." You've mixed up the loaders - that message is the image node complaining about your MP4. Swap in the video loader.

Playback looks fine but frames look wrong downstream. Because this node doesn't decode, the container/codec has to be something ComfyUI's video path can read. If a downstream frame loader chokes, that's a ComfyUI-side capability question, not this node's allowlist.

The clip is on the CDN but the graph says the URL is invalid. Same allowlist as the image loader - an Ace CDN HTTPS link, no other host, no redirect. Re-retrieve the task if the link has aged out.