Image to Video

The node that offloads the expensive part

ComfyUI's video stacks are brutal to run locally - Wan, Hunyuan, LTX all need serious VRAM and a patient workflow. The Animon Image to Video node takes a different route: it ships your first-frame image and a text prompt to Animon's hosted API, waits, and drops a finished video back into your graph as a native ComfyUI VIDEO. No local diffusion happens at all. If you don't own the GPU for anime-style video, or you just want a 1080P clip without babysitting a 30 GB model download, this is the node that makes cloud video feel like part of the graph instead of a detour to a website.

It's the flagship of the ComfyUI-Animon pack, and it's the one worth building a workflow around.

How it works

The pack is an OpenAI-compatible client in disguise. The Key node builds an openai.OpenAI client pointed at https://platform.animon.ai/api/openai/v1/, and this node uses it like any hosted video API: your IMAGE tensor is converted to PNG bytes, a videos.create call fires with the model, prompt, and your chosen resolution/seed/frame count, and then the node polls videos.retrieve every ten seconds until the job is completed. The downloaded file is decoded with decord back into a tensor, wrapped as ComfyUI's VIDEO type, and handed to you.

That last part is the elegant bit. Because the output is a real VIDEO tensor, it plugs straight into ComfyUI's core SaveVideo node - you don't need any extra file handling. The workflow included with the pack is literally LoadImage → Animon Image to Video → SaveVideo.

The inputs that matter

• model - anicut-1-5 (default), anicut-pro-1-5, or anicut-pro-1-6. The -pro- tiers are better quality; pro-1-6 is the only one that allows 1080P output. If you're aiming for HD, that's your pick.
• image - a single IMAGE tensor, usually from LoadImage or from your own SD pipeline.
• prompt - free text describing the motion and scene. Multiline.
• resolution - 480P (default), 720P, or 1080P. Check the capability map for what your model actually supports.
• num_frames - 17 to 81, in steps of 4, default 81. At 16 FPS that default is roughly five seconds of video.
• seed - default 42, with "control after generate" so you can reroll easily.

Outputs are VIDEO (the tensor, wire it to SaveVideo) and VIDEO_BYTES (the raw mp4 bytes - hang onto this one, it's the fast path for chaining an upscale later).

Installing and using it

ComfyUI Manager → Custom Nodes Manager → search ComfyUI-Animon → Install, then restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/IamCreateAI/ComfyUI-Animon.git
pip install -r requirements.txt

Then restart ComfyUI. Note the requirements are just openai and decord - no local model downloads. You do need a newer ComfyUI, since the pack uses the modern io.ComfyNode definitions, and you need an API key from the Animon platform. Key node first, wire it to this node, done.

Where people get burned

The biggest surprise is usually the billing one: every rerun on the seed is a paid API call, and generation plus the 10-second polling loop means a job takes a while. This is a per-generation service, not a local model - turn the node's seed to "control after generate" and you're spending credits on every reroll. And never share a workflow that still has your API key sitting in the Key node; strip it first or disable the key on the platform.