🅓 Frame Maker

The name makes it sound like it renders frames. It doesn't. FrameMaker is a compositor: it takes one subject image, pastes it over a background, and moves and scales it across frame_count steps to emit a batch of keyframes. No diffusion model involved, no weights to download, no API key, nothing running on your GPU - just PIL doing pan/zoom math while your VRAM sits idle.

It exists because animation in ComfyUI needs starting points. You can't get coherent motion from a single still, and hand-placing a character frame by frame is misery. FrameMaker generates the keyframes instead: you get a clean camera move (or subject move) on a transparent or cut-out subject, then hand those frames to something that adds motion - AnimateDiff with SparseCtrl, LayerDiffusion, or a frame-interpolator like RIFE VFI before VHS_VideoCombine turns the batch into a clip. That's exactly the pipeline diSty (diStyR on Reddit) built it for - his "splash" guided-motion workflow - and it's the shape to keep in your head when you wire this up.

How it works

Mechanically it's simple and old-school: your subject_image is converted to PIL, optionally joined with subject_alpha into an RGBA matte, and pasted onto a background at a position that linearly interpolates from x_pos_start/y_pos_start to x_pos_end/y_pos_end across frame_count frames. Scale does the same from scale_start to scale_end. That's the whole trick.

Two things quietly override your manual values, and both are in the source:

• movement_presets (25 of them, "left to center", "center to right", diagonals, the lot) replace x_pos_start/end and y_pos_start/end when it isn't None. movement_presets_distance is a percentage of the image dimensions, default 50 - so on a 768px canvas that's 384px of travel.
• movement_direction replaces your scale values outright: front sets scale 0.3 → 1.3 (subject grows toward camera), back sets 1.3 → 0.3. Set it to None if you want scale_start/scale_end to actually mean something.

scale_anchor_point (center, corners, top/bottom center) controls where the scaling pivots, and the optional bg1–bg4 backgrounds cycle frame by frame - handy for a multi-shot clip. Connect the frames input and it ignores the backgrounds entirely, using those frames instead. invert_order flips the output batch, which is a free reverse-play.

The inputs that matter

A beginner really sets about four things, everything else defaults sensibly:

• subject_image - your character/object, ideally already cut out.
• frame_count - 4 is the default and it's a good keyframe count; you'll interpolate up later.
• movement_presets + movement_presets_distance - the easiest way to get a clean pan.
• invert_alpha - flips the alpha matte; if your subject renders as a see-through ghost or a block, flip it.

Outputs are the money shot: frames (the IMAGE batch - wire it into your sampler, VFI, or video combine), plus subject_image, width, height, and frame_count ints for downstream sizing.

Install

The pack ships no requirements.txt and downloads no models - this is the rare custom node that's genuinely light.

cd ComfyUI/custom_nodes
git clone https://github.com/diStyApps/ComfyUI_FrameMaker
# restart ComfyUI

Or just use ComfyUI Manager and search "Frame Maker" - that's the recommended path in the README, and Manager will keep it updated.

Troubleshooting

• frame_count of 1 crashes - the interpolation divides by frame_count - 1, so a single frame is a divide-by-zero. Keep it at 2 or more.
• Your scale values get ignored - movement_direction is set to front or back. Set it to None.
• Your x/y positions get ignored - a movement preset is overriding them. Set movement_presets to None.
• Subject looks like a hole - flip invert_alpha.
• Output is 768×768 no matter what you loaded - resize_frame_input defaults to true and forces everything to frame_input_width/frame_input_height (768). If you want your real resolution, either change those or set resize_frame_input to false and keep subject and background the same size so the paste math lines up.