Agnes Video 2.0 I2V Config

Anchoring a video to a real image solves the consistency problem that plagues pure text-to-video - the character is who it is because the first frame says so. AgnesVideo20I2VConfig is the MediaHub node that packages that idea for Agnes's Video 2.0 API, and it's the more interesting of the two Agnes video configs because of its mode input. The provider supports three ways to use your images, and this node exposes all of them.

The mode input - the thing that matters most

• i2v - classic image-to-video. One source image becomes the first frame and the model animates forward from it.
• multi_image - multiple images feed the generation. Agnes treats them as references (or stages) rather than a single anchor, which is handy for steering a character or scene across a longer clip.
• keyframes - your images act as keyframes the model interpolates between, giving you more deliberate control over the arc of the motion.

Switching modes changes what the API does with your image input, and the config node passes that choice straight through to the payload. If you're coming from local I2V workflows, start with i2v; the other two are where this provider earns its keep.

Inputs that matter

• mode - i2v, multi_image, or keyframes (above).
• image - the source IMAGE. A batch of images becomes a batch of base64 JPEG data URLs, one per frame/reference slot depending on mode.
• prompt - what happens in the clip. With a strong first frame you can get away with a short prompt.
• negative_prompt - the usual unwanted-artifact filter.
• num_frames (default 121, max 441), frame_rate (default 24), num_inference_steps (default 30) - same controls as the T2V variant; they're your budget dials.
• width / height - defaults 1152×768, 64–4096 in steps of 8.
• use_seed / seed - deterministic-ish takes.

Output: request (MEDIAHUB_VIDEO_REQUEST) → MediaHubSubmitVideo → MediaHubResolveVideo → SaveVideo.

Installing it

Standard pack install:

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; the node is under MediaHub/Provider/Agnes/Video. Only requests, Pillow, and numpy are needed. You need an Agnes API key; nothing renders locally.

Common gotchas

The image-to-API conversion is the part that quietly bites. Source images get JPEG-encoded before upload, so extremely detailed or text-heavy frames will lose a little fidelity - usually fine for animation, occasionally a problem if your keyframe contains readable text. And if you pick multi_image or keyframes but only wire in a single image, expect the provider to do something unexpected; those modes are built around having multiple references. The pipeline mistakes (no Resolve, no key) are the same as every other video config in this pack.