⚠ Load Batch From Folder (Advanced)
Skip this one — it's marked deprecated, and the replacement is better
- images
- masks
This article is short on purpose. Load Batch From Folder (Advanced) is a deprecated node - it literally ships with a ⚠ in its display name, sits in the Eclipse/ Legacy category, and its own description says "DEPRECATED: Please use Load Batch From Folder (Step Advanced) instead." If you're reading this because you googled the node, the most useful thing I can tell you is: don't build anything new on it.
Why it exists and why it's gone
In earlier versions of the pack, this was the advanced batch loader: it pulled every image from a folder (or several, one path per line, with recursion) into a single batch, then normalized sizes with a full arsenal of options - resize_mode (first/largest/smallest/none/list), scale_to, custom W×H, aspect-ratio overrides, fit modes from resize and crop through pad, pad_edge, pillarbox_blur and stretch, crop anchoring, padding color, resampling method, divisible_by rounding, and a device picker for resize ops (with the note that lanczos isn't supported on GPU and falls back to bicubic). That's a genuinely capable node.
Then the pack hit its 4.0 cleanup - the README is explicit that all legacy/deprecated nodes were removed from the codebase - and the "Advanced" loader was superseded. The replacement, Load Batch From Folder (Step Advanced), keeps the entire resize arsenal and adds the two things the old one lacked: a batch_index that shifts the frame window forward so you can walk through a folder in fixed-size batches, and a files output carrying the absolute paths of what was loaded. Same heavy machinery, plus stepping, plus filenames. There is no reason to prefer the deprecated version.
What to do if it's already in your workflow
If you load an old workflow that references this node and it fails, that's the version-tag removal the README warns about. Two options: drag the replacement in and rewire (the inputs map across almost 1:1), or - if you have a large legacy workflow - use the pack's built-in Workflow Migration Tool node or the python tools/migrate_workflow.py script, which upgrades saved workflows from legacy Eclipse node versions to 4.x automatically (with backups). Don't hand-migrate a workflow with dozens of these.
Install
It comes with ComfyUI_Eclipse:
cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI_Eclipse
But again - install it, and then use the Step Advanced version that lives in the same pack under Image/Loaders. This one's a dead end; the successor is the same thing with stepping and file paths on top.
Inputs (18)
| Name | Type | Default | Description |
|---|---|---|---|
| folder_path | STRING | Path(s) to folder(s) containing images. One folder per line. Absolute paths are supported anywhere on the filesystem. All images are loaded and returned as a single batch. | |
| include_subfolders | BOOLEAN | false | Include images from subfolders recursively. |
| sort_by | COMBO | name | How to sort images within each folder. |
| sort_order | COMBO | ascending | Sort order for the image list. |
| frame_start | INT | 0-99999–99999 | First frame to include from the combined frame list. 0 = first frame. Negative values count from the end (-1 = last frame). |
| frame_end | INT | -1-99999–99999 | Last frame to include, inclusive. -1 = last frame. Negative values count from the end. Range is applied before resize_mode normalization. |
| resize_mode | COMBO | first | How to handle images with different sizes. 'first': resize all to the first image's dimensions; 'largest': resize to the largest W×H found; 'smallest': resize to the smallest W×H found; 'none': raise an error if any sizes differ; 'list': skip stacking entirely and return images as a list — allows mixed sizes, useful for preview testing; 'custom': resize using the advanced sizing and aspect ratio rules below. |
| scale_to | COMBO | longest | Which dimension to constrain: longest side, shortest side, width, height, total pixels (kilo-pixels), or custom W×H. |
| size | INT | 10241–16384 | Target size for the chosen scale_to mode. For total_pixels this is in kilo-pixels (e.g. 1024 = ~1M pixels). |
| custom_width | INT | 5120–16384 | Target width when scale_to is 'custom'. 0 = keep original width. |
| custom_height | INT | 5120–16384 | Target height when scale_to is 'custom'. 0 = keep original height. |
| aspect_ratio | COMBO | original | Override aspect ratio. 'original' keeps the input image ratio. |
| fit | COMBO | resize | How to fit the image into target dimensions: • resize — scale proportionally (output may be smaller than target) • crop — scale to fill then crop excess • pad — scale to fit then pad with solid color • pad_edge — pad with mean color of nearest edge • pad_edge_pixel — pad by replicating edge pixels outward • pillarbox_blur — pad with blurred, desaturated, darkened background • stretch — distort to exact target size |
| crop_position | COMBO | center | Anchor point for crop and pad operations. |
| pad_color | STRING | #000000 | Background color for pad mode (hex, e.g. #000000). |
| method | COMBO | lanczos | Interpolation method for resampling. |
| divisible_by | INT | 81–512 | Round output dimensions to nearest multiple of this value. |
| device | COMBO | cpu | Device for resize operations. GPU is faster for large images. Lanczos is not supported on GPU and falls back to bicubic. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | Loaded (and optionally resized) images batch or list. |
| masks | MASK | Corresponding masks batch or list. |