Create CameraCtrl Poses (Combo) 🎭🅐🅓②

CameraCtrl is the most setup-heavy feature in this whole pack, and this node is how you build the actual camera move for it without hand-authoring a pose file. It's displayed as Create CameraCtrl Poses (Combo), and its whole job is letting you pick up to six camera-motion segments from dropdowns, set a speed and a frame length, and get out a CAMERACTRL_POSES object you feed into a CameraCtrl workflow.

Before you reach for this node at all, know what CameraCtrl actually requires, straight from the README: the specific pruned model CameraCtrl_pruned.safetensors, an AnimateDiff SD1.5 motion model - specifically trained for v3 - and it's Gen2-only, with its own helper nodes grouped under the Gen2/CameraCtrl submenu that this node belongs to. None of the Gen1 loaders support it. If any of those pieces are missing, this node's output won't do anything useful downstream.

How it works

Camera control here works by composing a sequence of named camera movements - think pan, tilt, zoom, roll, in some direction and magnitude - into a combined pose trajectory the CameraCtrl model can follow during sampling. Rather than one single movement for a whole clip, you can chain up to six motion segments in this node alone via motion_type1 through motion_type6, and the prev_poses input lets you chain multiple ADE_CameraPoseCombo nodes together for even longer, more complex camera sequences.

The inputs and outputs that matter

Required: motion_type1 through motion_type6, each its own dropdown picking a camera movement for that segment of the sequence - leave a slot on its default/neutral option if you don't want that segment doing anything. speed (default 1, range −100 to 100 - negative values presumably reverse the movement direction) and frame_length (default 16) round out the segment's timing.

Optional: prev_poses (type CAMERACTRL_POSES), which chains this node's output onto a previous ADE_CameraPoseCombo's poses, letting you build a longer camera sequence across multiple nodes rather than cramming everything into six slots. Output is CAMERACTRL_POSES, which only feeds into the CameraCtrl-specific Apply nodes under the same Gen2/CameraCtrl submenu.

How to install it

Standard for the pack - ComfyUI Manager, search AnimateDiff Evolved by Kosinkadink, or:

cd ComfyUI/custom_nodes && git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved

Then, specifically for CameraCtrl: download CameraCtrl_pruned.safetensors from huggingface.co/Kosinkadink/CameraCtrl (linked in the README) and an SD1.5 v3 motion model (v3_sd15_mm) - CameraCtrl was trained specifically against v3, not v1 or v2.

Common issues & troubleshooting

Camera movement doesn't show up at all, or looks nothing like what you configured. Check the fundamentals first: are you on a Gen2 graph (Gen1 loaders don't support CameraCtrl at all), is your motion model specifically v3_sd15_mm, and is the CameraCtrl model the pruned version the README links rather than some other CameraCtrl checkpoint you found elsewhere.

Not sure what each motion_type dropdown option actually represents. These are populated by the node's own combo list rather than documented individually in the README's prose - the safest way to learn what each does is to build a short test clip with one segment at a time and watch what changes, rather than guessing from the name alone.

Built a long sequence and it feels disconnected between segments. That's what prev_poses chaining is for - if you're only using the six slots on a single node and want a smoother longer sequence, chain a second ADE_CameraPoseCombo node in via prev_poses rather than trying to cram more into one node's limited slots.