IAMCCS_WanImageMotionPro

IAMCCS_WanImageMotionPro is the grown-up version of the WanImageMotion motion node, and the headline feature is right in the schema: an end_samples input and an end_lock_slots control that hard-lock the last latent slots to a target end frame - FLF-style first-last-frame continuity, on top of the motion amplitude boost. If you're building shot-to-shot video where every chunk is supposed to land on a specific final frame, this is the node that makes that promise stick.

It's a big node, honestly - 30+ inputs, a preset dropdown, and several families of controls. But the tiers sort themselves out:

The controls that earn their keep

• motion / motion_mode / motion_latent_count / safety_preset - inherited from the base WanImageMotion, same behavior. The amplitude fixer upper.
• use_end_frame + end_lock_slots - the new core. With end_samples connected, end_lock_slots = 1 locks the very last latent slot to the end image so the final frame lands exactly on target. This is what turns "continuation" into "a planned A→B transition."
• end_overshoot_slots - the trick that avoids the hard freeze. It extends the internal generation window by a few latent slots and locks the end image at the tail of the extended range, so the visible clip converges toward the end frame instead of sitting on it. Then you cut the overshoot with the trim_slots output. Recommended 1 (+4 frames), 0 disables.
• clip_vision_mode - how start/end CLIP vision embeddings are merged. auto (concatenate both, the original FLF behavior) is correct only for true A→B transitions. For SVI Pro continuation where anchor and end are the same image, use end_only - it prevents crossfade bleed. This is a genuinely important switch for people chaining SVI clips.
• prev_samples_profile / prev_skip_last_slots / prev_tail_pick_mode / prev_anchor_blend - the "where do I pick the motion tail from" family, for when your prev tail comes from a non-FLF generator or a slightly earlier motion state. mixed_svi_safe ignores the last prev slot and uses the latest usable tail - the go-to for SVI-to-Pro chains.
• latent_refresh / delta_max - AdaIN-style latent moment stabilization, same as the base node.
• continuity_profile - raw-like continuity paths for SVI chains; most people can leave this off and use the explicit controls.
• raw_mode - strict 1:1 parity with the original WanImageToVideoSVIProFLF behavior. Enable it to benchmark, disable for real work.
• preset - server-side presets (raw, svi continuity, flf overshoot, parity 1:1, ...). Leave [custom] if you've set widgets by hand.

Outputs: positive, negative, latent (feed the sampler), and trim_slots (an INT telling you how many slots the overshoot added - wire it into IAMCCS_WanPrevTailPrep to clean the tail for the next segment).

Install

Same pack, same story:

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

or ComfyUI Manager → search "IAMCCS" → restart. No model downloads; the pack floor is ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

The gotchas worth memorizing

end_transition_frames is deprecated - keep it at 0. Values above 0 blend padding latents toward the end frame in latent space, and because VAE latent interpolation isn't linear you get static, frozen output. That's the author's own warning in the tooltip, and it's accurate.

Second: if you have KJNodes WAN attention optimizations enabled (FETA / wrapped attention), some versions crash with an einops shape mismatch when end_overshoot_slots > 0. The documented workaround is to set overshoot to 0 or update/disable that KJNodes optimization. Annoying, but knowing it saves you an afternoon of staring at a traceback.

Finally - this node's latent output is sized for sampling, so check length is in step with your WAN frame-count conventions (81 by default) or you'll encode a clip that's not the length you think.