The help pages, the README and the changelog use the new names (Round canvas to, Frames for, Server file, the Resize switch) and describe the always-scrubbing timeline and its grip, the padlock rules, the "already this shape" note, the resize warning and Even out, the preview switch bringing its picture back, Compare's labels and the Workflow Note's Use it.
3.5 KiB
Video Crop + Rotate + Pad → Frame
Outputs one frame, not a clip. The timeline in the editor is for finding the frame; the node returns that single image, not a re-encoded video.
Loads one exact video frame and applies the same rotate → crop → pad
transform as the image node. It returns the transformed image, its
generated-area mask, a stitcher for Stitch Inpaint, the selected
original frame before the transform, and the output width and height.
Source and frame
- Uploads: Pick an input video (the list plays the clip under the pointer, muted, and filters as you type), click Upload in the compact source card, or drop a video file onto the node. Another video keeps the fill, feather and resize settings and the lit format chip (the new frame uses its Crop / Pad mode); rotation, crop and the playhead start over.
- Canvas row under the format chips: fill swatch, feather amount and the Resize off | on switch on the node face; turning Resize on opens the megapixel budget and its Step. The line under the picture names each step that sets the output size, as on the image node.
- Upload streams to the input folder, so videos larger than the buffered image-upload limit can be selected here or dropped onto the node.
- Uploads / Server file (
source_mode:input folder/local path): the source card's switch. Server file mode avoids copying large files. - local_path: Absolute path used only in Server file mode, read in place without an upload copy. It must point inside ComfyUI's input, output or temp folder, for queued runs and the editor's live preview alike; paths anywhere else are refused.
- Playhead: The rail under the node's preview is the timeline: press or drag it to
scrub, or type a frame in the Frame box; the frame on the stage is the frame the
node outputs (
frame_index, withframe_timekept in step). - seek_mode / frame_time: Time-based seeking for graphs that drive the position by link; the editor itself always works in frames.
The editor has the same rail full width, plus first/last, ±1, ±25, ±50, ±100, play/pause, keyboard arrows and Home/End. Only one frame preview request remains active; stale requests are cancelled, and frames already shown are served from a small cache.
Transform and outputs
Rotation, crop, padding, feathering, fill, canvas multiple, the megapixel resize, gestures, image, and mask match the image node. The output batch contains one frame.
The handles work on both the node and the editor. The ratio buttons, Fit, the padlock and the orientation button work as on the image node: a tap pads (or, with Fit on crop, crops) the frame to a ratio, a lit ratio is the shape the canvas has now, and the padlock keeps that shape while you drag (the row says Held). Reset crop restores the full crop while keeping rotation and padding. Reset clears rotation, crop and padding and turns the padlock off; fill, feather, Round canvas to and the playhead stay. Round canvas to sets the canvas pixel multiple, with 1 turning it off.
The appended stitcher output restores kept pixels after generation with a 32-pixel blend. original returns the chosen RGB frame before the transform, and width / height the output size after the transform and any resize. The existing image and mask sockets keep their positions.
Video metadata and preview routes validate extensions and return only the information needed by the editor. The node performs no remote network requests and does not rewrite the source video.