Image Motion Influence

One thing to know before you even search for this node: the class name has a typo baked in. It's ImageMotionInfluance, not "Influence" - the pack author's own README calls it "ImageMotionInfluence" in the feature list, but the actual node in your graph is misspelled. If the node search box turns up nothing when you type the correct English word, that's why.

Typo aside, this is a genuinely handy little utility: feed it one still image and it spits out a sequence of frames that fake a camera pan and zoom, no diffusion model, no VRAM budget, no waiting. It's the Ken Burns effect as a ComfyUI node. Reach for it when you want a quick motion preview before spending real GPU time on an actual video model, or when you just need a simple parallax-style clip and a full I2V pipeline is overkill.

How it works

It's pure image transformation - no model runs, no checkpoint loads. Across frame_num output frames, it progressively shifts the image horizontally (the pan) and scales it in (the zoom), then, per the pack's own description, uses mirror techniques to close the sequence into a seamless loop rather than cutting hard at the end. If your output seems to "bounce back" partway through, that's the loop mechanism working as designed, not a bug.

The inputs and outputs that matter

• image - your source still.
• move_range_x (default 0, range -150 to 150) - horizontal pan distance in pixels. Sign flips direction. This is a straight pixel-space shift within your source image, not a learned outpaint, so a wide range on a tightly-cropped photo will run out of real image to show and start looking rough at the edges. Start conservative - 30 to 50 - and only push higher if your source has margin to spare.
• frame_num (default 10, range 2–500) - exactly how many frames come out. There's no interpolation beyond what you ask for, so match this to your target clip length at your export fps, or the motion plays at the wrong speed.
• zoom (default 0, range 0–0.5, step 0.05) - progressive zoom-in strength. 0 is a pure pan with no scaling.

Output is a batch of IMAGEs. Feed it straight into a video-combine/save-animated node for a standalone clip, or use it as a starting sequence before an actual I2V model. If you're heading toward this pack's HunyuanVideoSamplerSave, note that node wants a LATENT, not an IMAGE batch - you'd still need to resize and VAE-encode along the way, so pair this with ResizeImageForHunyuan first if dimensions matter downstream.

How to install it

Via ComfyUI Manager: search ComfyUI-HunyuanVideoSamplerSave, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ShmuelRonen/ComfyUI-HunyuanVideoSamplerSave.git

then restart. This is the one node in the whole pack you can use with zero downloads - no Hunyuan model, no extra dependencies, it never touches the model referenced elsewhere in the README.

Common issues & troubleshooting

Motion looks jerky or mechanical. It's a linear pixel-space transform, not a learned motion model - there's no scene understanding, no parallax between foreground and background. Dial the range and zoom down for subtlety; this is a blunt instrument, not a substitute for a real I2V model's camera control.

Edges look torn up at high move_range_x. You're panning past the actual content of your source image. Either use a source with more margin around the subject or pull the range back.

The clip plays too fast or too slow. frame_num is a frame count, not a duration - check it against the fps your downstream save/combine node is set to.

Can't find the node by name. Search "Motion" or "Hunyuan" in the node picker instead of typing "Influence" - the class name is ImageMotionInfluance, spelled exactly like that.