Start-End to Video

The node for "I know how it starts and how it ends"

Image to Video in this pack gives you motion from a single frame and a wish. Start-End to Video is the more controllable sibling: you give it a starting frame and an ending frame, and Animon's anicut-ib models (the ib is image-begin/image-end) generate the in-between. Think character turning toward the camera, a camera push-in, a scene that needs to land on a specific final shot instead of wherever the model decided to wander.

It's the pack's interpolation node, and for anime-style keyframes it's the one you reach for when a plain text prompt can't pin down the destination.

How it works - and the one thing that's different

Mechanically it's the same cloud pipeline as the rest of the pack: the Key node's OpenAI-compatible client calls videos.create with your model, prompt, resolution, seed, and frame count, the node polls videos.retrieve every ten seconds until the job completes, and the result is decoded with decord into a native ComfyUI VIDEO tensor plus VIDEO_BYTES.

The critical difference is in the inputs: image_start and image_end are ANIMON_IMAGE_IDs, not image tensors. You can't drop a LoadImage output straight in. Those IDs only come from an upload node - Upload Image (for files in ComfyUI's input folder) or Upload Image (from Tensor) (for images your own SD pipeline just produced). The README is explicit: if you're using more than one image, you must upload them to Animon's server first. So the wiring is always Upload Image → Start-End to Video, and only then does the generation node exist.

The README also hands you a genuinely useful tip: start and end frames can be the same image. When they are, you don't need a second upload - reuse the IMAGE_ID from a previous generation instead of uploading again, and use the node like an animated version of that one still.

The inputs that matter

• model - anicut-ib-1-5 (default), anicut-pro-ib-1-5, or anicut-pro-ib-1-6. The pro-1-6 tier is the only one that unlocks 1080P.
• image_start / image_end - the uploaded IMAGE_IDs. Both required.
• prompt - optional-ish; describes what should happen between the frames. Multiline, defaults to empty.
• resolution - 480P (default), 720P, 1080P. Mind the capability map - the base 1-5 models top out at 720P.
• num_frames - 17 to 81 in steps of 4, default 81. That's about five seconds at Animon's 16 FPS.
• seed - default 42, with control-after-generate for easy rerolls.

Outputs are VIDEO (wire to ComfyUI's core SaveVideo) and VIDEO_BYTES (the raw mp4, useful for chaining an upscale without re-encoding).

Installing and the gotchas

Same story as the whole pack: ComfyUI Manager → search ComfyUI-Animon → restart (or git clone into custom_nodes plus pip install -r requirements.txt). Recent ComfyUI required for the io.ComfyNode definitions, API key from the Animon platform required for everything else.

The trap most people hit is forgetting the upload step entirely and wondering why the node's inputs are grayed out - there's no image widget here because there's no tensor. And each generation is a paid cloud call: every seed reroll spends credits, and unlike a local model there's no "try this free." Set your key once in the Key node, keep it out of any workflow you share, and let the interpolation do the aiming.