Load Plate:
Load Plate brings original footage into a Paint a plate workflow from a video file or a sequence of still frames, including EXR, PNG, TIFF, and DPX. A “plate” is the original shot you want to edit while preserving its visual character and delivery format. The node is designed to load footage at full precision and retain the source information needed by Save Plate, helping you carry the shot through editing and back toward its original format. Videos are decoded using ComfyUI’s video loader, while image-sequence files are decoded in parallel to speed up loading. The node preserves the plate’s existing colour space rather than automatically converting its appearance, so colour conversion should be handled separately with Color Space (OCIO). It also supports repeating the final frame to make a shot’s length compatible with LTX while keeping track of how many frames were actually loaded.
Load Plate Input Parameters:
path
A required STRING identifying the footage to load. You can specify a video file, a folder containing an image sequence, or a numbered frame pattern such as shot.####.exr. Paths are resolved through the extension’s plate-path handling, with the node’s interface directing you to use footage in ComfyUI’s input folder. Start the path with output: to read from ComfyUI’s output folder instead; for example, output:shot.####.exr. This setting determines which source is inspected and decoded, including the source information that accompanies the loaded footage. No explicit default value, length limit, or fixed list of choices is defined in the node schema. Use the exact folder, filename, or frame pattern for your footage, and keep image sequences in clearly organized folders so the intended shot is selected.
max_frames
A required INT controlling the maximum number of source frames loaded from the beginning of the footage. The default is 121, the minimum is 1, and the maximum is 1024. For a video, the node requests a decoding duration calculated from this limit and the detected frame rate, then limits the resulting image batch to this number of frames. For an image sequence, it selects up to this many files before decoding them. If the source contains fewer frames, the node returns the available decoded frames rather than creating additional source content. Lower values reduce loading time and memory use, which is useful for testing a workflow before processing a longer shot. At 24 fps, 121 frames represent approximately five seconds. This limit applies before optional LTX padding, so the final image batch can be longer when padding is enabled.
pad_8n1
A required BOOLEAN with a default of False and options True or False. When True, the node repeats the last loaded image and its alpha until the batch length reaches a number of the form 8k + 1, which is the frame-length pattern required by LTX. For example, 120 loaded frames become 121, while 122 become 129; a batch already containing 121 frames needs no extra copies. Padding does not create new motion or decode additional footage—it adds a held final frame. The reported original frame count remains unchanged, allowing you to distinguish actual loaded footage from compatibility padding. Leave this disabled when your next processing step does not require this frame-length pattern or when you want the image batch to contain only the frames actually loaded.
Load Plate Output Parameters:
images
An IMAGE batch containing the decoded frames in source order, limited by max_frames before any optional padding. This is the visible footage you connect to image-processing, editing, or colour-conversion nodes. If pad_8n1 is enabled, the batch may end with repeated copies of the final decoded frame. The node transfers the resulting batch to ComfyUI’s configured intermediate processing device and data type. It does not perform an OCIO colour conversion: the pixels remain in the plate’s existing colour space, so a source that requires conversion may not look correct in a display-oriented preview until you convert it appropriately.
alpha
A MASK batch containing the source alpha when it is available. Alpha describes opacity: values of 1 indicate opaque areas, values of 0 indicate transparent areas, and intermediate values represent partial transparency. If the decoded source has no alpha, the node creates an all-ones mask at the image dimensions, meaning the entire image is opaque. When frame padding is enabled, the last alpha frame is repeated alongside the last image frame so both batches remain synchronized. This output represents source transparency, not an automatically generated selection of an object or an edited area; check the mask convention of the next node before using it for compositing or inpainting.
plate
A PAINTAPLATE_PLATE object containing the source specification used within the Paint a plate workflow. This is structured source information rather than another image batch, and it is intended to preserve the information Save Plate needs when writing the edited shot. Before returning it, the node updates the specification with the actual loaded frame count and a colour-space guess. Keep this output available for compatible downstream nodes when you want the workflow to retain the original plate’s format information. The colour-space guess is a suggested interpretation of the source, not a colour conversion applied to the images.
fps
A FLOAT giving the frame rate recorded in the probed plate specification. This describes the intended playback speed in frames per second and is useful when passing the shot into timing-sensitive processing or export steps. For example, a value of 24.0 means that 24 frames represent one second of playback. Padding does not change this value; it only changes the number of frames in the image and alpha batches. For image sequences, the value comes from the plate-probing system, so verify that it matches your intended shot timing rather than assuming individual still-image files establish a playback rate.
frames
An INT giving the number of frames actually loaded before optional padding. It reflects the decoded footage available within the requested limit, not necessarily the total length of the source. For example, if you load 120 frames and enable padding, this output remains 120 even though images and alpha contain 121 frames. This distinction is useful when you need to identify the original content length or remove repeated end frames later. Without padding, this count matches the length of the returned image batch.
spec
A STRING containing a JSON representation of the plate specification. It provides a readable or transferable version of the source information, including the loaded frame count and the colour-space guess added during execution. You can use it to inspect how the footage was identified and to help diagnose unexpected timing or colour interpretation. The exact additional fields depend on the source and the plate-probing implementation. This text output is not interchangeable with the structured plate output when a downstream node requires the PAINTAPLATE_PLATE type.
Load Plate Usage Tips:
- Begin with a small
max_framesvalue, such as 9 or 17, to check decoding, colour interpretation, and downstream connections before loading a longer shot. - Use
pad_8n1=Truefor an LTX workflow that requires that frame-length pattern. Values such as 121 already fit the requirement, so no end-frame repetition is needed at that length. - Keep
pad_8n1=Falsewhen you need an exact batch length without repeated frames. When padding is enabled, useframesto track the real loaded length rather than assuming every returned frame is original footage. - Use an explicit sequence pattern such as
shot.####.exrwhen identifying numbered footage, and keep unrelated images in separate folders to reduce ambiguity. - Use the
output:prefix when loading footage saved under ComfyUI’s output folder instead of placing an output-relative path into an input-folder lookup. - Add Color Space (OCIO) when your source needs conversion for the next processing stage. Confirm the source colour space rather than relying solely on the reported guess, especially with EXR or other production footage.
- Preserve the
plateconnection for the Save Plate stage, and usespecto inspect the source information when troubleshooting. - For high-resolution footage, reduce
max_framesfirst if loading becomes slow or memory use is excessive. Sequence decoding is already parallelized internally; this node does not expose a worker-count setting.
Load Plate Common Errors and Solutions:
File not found or source path cannot be resolved
- Explanation: The footage cannot be located using the supplied
path. This can happen because of a spelling mistake, an incorrect folder, a mismatched sequence pattern, or a missingoutput:prefix. The supplied class does not define an exact error message for this condition; the wording comes from the underlying path or file-loading code. - Solution: Confirm that the file or folder exists, check capitalization and frame-number formatting, and use a path relative to the intended ComfyUI folder. Add
output:when reading from the output folder.
No matching sequence frames or no decoded images
- Explanation: A folder or frame pattern may not identify any usable frames, or decoding may produce no images. The node expects at least one decoded component and a usable image batch; an empty source can therefore cause a later indexing or concatenation failure. Exact error wording depends on where loading fails.
- Solution: Check that the sequence folder contains readable images and that the filename pattern matches their numbering. Try loading a known-good individual frame to separate a sequence-selection problem from a decoding problem.
Unsupported format or video decoding failure
- Explanation: The underlying loader may be unable to open a file because its encoding is unsupported, the file is damaged, or its contents do not match its extension. The class does not provide a fixed decoder error string.
- Solution: Test the file in another application, confirm that the copy is complete, and try a known-good file of the same intended format. If necessary, re-export the footage in a format and encoding supported by your ComfyUI installation.
Permission denied
- Explanation: ComfyUI may be able to identify the source path but lack permission to inspect or read the footage. Access is needed both when checking source modification times and when decoding files.
- Solution: Give the account running ComfyUI read access to the source folder and its files, or copy the footage into an accessible ComfyUI input folder. Check access to mounted or network storage if applicable.
Image sequence frames have incompatible dimensions or channels
- Explanation: The node concatenates decoded frames into a single batch, which requires compatible image dimensions and channel layouts. Mixed resolutions or inconsistent image formats can cause tensor-size errors. If the first decoded component has alpha, the code also expects alpha to be available in the remaining components.
- Solution: Make sure all sequence frames use the same resolution and compatible channel configuration, including consistent alpha availability. Remove unrelated images from the folder or re-export the sequence consistently.
Out of memory or allocation failure
- Explanation: Decoded footage can require substantial memory, particularly at high resolution or with long batches. The node holds decoded frames and combines them into batches, while padding and device transfer can require additional allocations.
- Solution: Reduce
max_frames, test a shorter section, close other memory-intensive workflows, or prepare a lower-resolution working copy when appropriate. Disable padding if the next stage does not need it.
Frames appear too dark, too bright, or washed out
- Explanation: This is not necessarily a loading failure. The node preserves the source colour space and does not automatically convert footage for viewing or model processing. A linear or otherwise differently encoded source may therefore look unexpected.
- Solution: Check the source’s documented colour space and apply the appropriate Color Space (OCIO) conversion. Inspect
specfor the colour-space guess, but confirm it before choosing a conversion.
Alpha is completely white
- Explanation: An all-white alpha output is expected when the decoded source has no alpha channel: the node supplies a fully opaque mask. It does not automatically isolate subjects or create an edit mask.
- Solution: If transparency should exist, confirm that your source export actually contains alpha and that the loader exposes it. If you need a subject or edit selection, create that mask separately.
Returned batch contains more frames than frames reports
- Explanation: This is expected when
pad_8n1repeats the final frame. The image and alpha batches include padding, whileframesand the specification’s frame count retain the number loaded before padding. - Solution: Disable
pad_8n1if you do not need LTX-compatible length, or trim the processed result back to the reportedframescount at the appropriate stage.
