WanImageMotion

Slow-motion is the classic SVI complaint: you chain a Wan continuation and the output drifts along at half speed, all floaty and listless. IAMCCS_WanImageMotion is a drop-in replacement for the common WAN SVI Pro image-to-video nodes that adds a motion amplitude control - crank it and your clip actually moves. It's the pack's motion-control workhorse, and its CTR on the site suggests it's one of the nodes people actually keep finding.

How it works

This is a conditioning node, not a sampler. You feed it positive/negative conditioning plus anchor_samples (the encoded start image latents), and optionally prev_samples (the previous segment's output latents for continuation). It then edits the conditioning's latent image in latent space: it takes the difference between each latent frame and the anchor, scales that difference by your motion gain, and writes it back - so frames that were barely deviating from the start image now deviate more, and the sampler produces a clip with real movement. A safety_preset guards the result: safe (default) applies per-channel stabilization, a tanh limiter, and a 2-frame ramp so you don't blow out colors at high motion; safer tightens that for motion > 1.5; base is 1:1 with the original FLF/SVI node; legacy mirrors the PainterI2V hard-clamp approach.

The settings a beginner actually touches:

• motion (1.0–2.0, default 1.15) - the amplitude. 1.15 is the gentle boost; 1.5+ is where you get obvious movement and should reach for safer.
• motion_mode - motion_only (prev_samples) boosts just the continuation tail; all_nonfirst (anchor+motion) boosts every frame after the first.
• vram_profile - normal, chunked_blocks_2/4, loop_per_frame, or cpu_offload for tight VRAM.
• latent_refresh (0–1) - matches per-channel mean/contrast of the motion tail to the anchor. Start at 0.5 for chains of 3+ segments; it fights the drift that makes long SVI clips wander.

Outputs are the edited positive, negative, and an empty latent ready for sampling. Wire the conditioning and latent into your Wan sampler and it works like the node it replaces.

The honest tradeoffs

More motion isn't free. Push motion high on a 14B offloaded model and you're paying for extra compute on every frame plus risking seam degradation - that's what the safety presets are for. And prev_samples is lazy and loop-friendly: on the first segment of a chain it may ask ComfyUI for a second pass, which is normal, not a hang. There's also a use_prev_samples toggle so you can keep the wire connected (autolink forces it sometimes) while telling the node to ignore it.

Install

Part of IAMCCS-nodes:

• ComfyUI Manager → search "IAMCCS" → install → restart.
• Or:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart ComfyUI. Requirements: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8. It rides on your existing WAN 2.2 + SVI setup - the pack's docs (docs/WanImageMotion.md) cover the full chain if you want the reference workflow.