AIHub Expose Project Image Batch

AIHub Expose Project Image Batch is the storage-side version of AIHubExposeImageBatch: it reads a series of images that were previously stored in the project and packs them into a batched IMAGE tensor (plus MASKS, WIDTH, HEIGHT). The README's guidance is blunt and worth quoting: "Use this whenever while using an action to create an image you used append" - in other words, this node is the read end of the append-based batch actions. If a workflow wrote frames with AIHubActionNewImageBatch (or called AIHubActionNewImage in APPEND mode repeatedly), this is how a later workflow gets that whole batch back.

Mechanically it works like every project node: the client resolves the stored files, the server injects their paths, and this node loads each one through ComfyUI's image loader before stacking them. Two details make it more useful than a plain batch read.

First, indexing. The indexes input is a comma-separated list with range support: 0,1,2 loads the first three images, 0-4 loads the first five, and negative indexes work, so -1 grabs just the last frame. That's your tool for "only re-process frames 5 to 8" without loading everything.

Second, normalization. Batches of mixed sizes will break a video model. The node's optional normalizer input takes an AIHUB_NORMALIZER from AIHub Utils New Normalizer; leave it empty and it defaults to targeting the image with the most megapixels using a nearest-exact upscaler. That's a reasonable floor, but for video work you'll usually want to set an explicit width/height.

The inputs that matter

• file_name - the batch filename without the number counter (e.g. exposed_image_batch.png); the client appends the per-file numbers.
• indexes - which images to load (see above).
• normalizer - optional; ties to the New Normalizer utility.

Plus id. Outputs are IMAGE, MASK, WIDTH, HEIGHT.

Install & gotchas

Standard pack install: ComfyUI Manager → search ComfyUI-aihub-workflow-exposer, or git clone https://github.com/otavanopisto/ComfyUI-aihub-workflow-exposer into custom_nodes, then restart. No requirements.txt, no model downloads.

The gotcha that bites most people: file_name is the stem - the number counter is added by the client, and you must give the base name without it or the files won't match up. Second: if the batch was stored with REPLACE (or the writing workflow never ran in append mode), there may be exactly one file and indexes pointing past it will find nothing. If the node complains "no valid images found," first check that the writer actually stored the files as a numbered batch in the project, not that the node is broken.