Agnes Video 2.0 T2V Config

Video generation from text is a heavy lift locally - that's the whole reason cloud APIs like this exist. AgnesVideo20T2VConfig builds a text-to-video request for Agnes's Video 2.0 model, and it's the most "local-workflow-like" of the video configs in the pack, because Agnes exposes frame counts, frame rates, and inference steps the way a ComfyUI user expects to see them.

As with every config node here, it produces a MEDIAHUB_VIDEO_REQUEST, not a video. The chain is: this config → MediaHubSubmitVideo (fires the job and returns a task id) → MediaHubResolveVideo (polls until the render finishes, downloads the MP4) → ComfyUI's SaveVideo. Skip any link in that chain and you have a request going nowhere.

Inputs that matter

• prompt - what you want to see. The whole film comes from this string.
• negative_prompt - yes, this provider still has one. Defaults to empty; useful if you keep getting the same unwanted artifacts.
• width / height - defaults 1152×768 (roughly 3:2), both 64–4096 in steps of 8.
• num_frames - default 121, max 441, step 8. This is the length of your clip in frames. At the default 24 fps that's about 5 seconds; crank it toward the cap and you're asking for a much longer (and more expensive) render.
• frame_rate - default 24, range 1–60. This and num_frames together determine duration, so don't fight them - if you want a longer clip, add frames rather than dropping the frame rate.
• num_inference_steps - default 30, range 1–100. More steps usually means cleaner output and slower, more expensive renders. 30 is a fine starting point.
• use_seed / seed - reproducibility. use_seed defaults to off; flip it on and give a seed to get the same-ish take on a prompt.

Output: request (MEDIAHUB_VIDEO_REQUEST) → MediaHubSubmitVideo.

Why you'd reach for it

If you want a video without burning your GPU for an hour (or without a GPU at all), this is the point-and-shoot option inside ComfyUI. It's also a low-commitment way to test whether Agnes's Video 2.0 is worth paying attention to - one render tells you more than any benchmark.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI and look under MediaHub/Provider/Agnes/Video. Dependencies: requests, Pillow, numpy. No models, no downloads - Agnes needs an API key and renders everything server-side.

Common gotchas

The config node itself is low-risk; the mistakes live in the pipeline. Forgetting Resolve is the big one - without it you get a job ID and no video. And remember video renders take minutes, so MediaHubResolveVideo will block your queue while it polls (up to ten minutes). If a render times out, the job may still finish on Agnes's side; check raw_json and the provider console rather than assuming it failed.