Load Result Images

Searching finds paths; this node finds images. Load Result Images takes a SEARCH_RESULTS object and actually loads the matching files off disk as full-resolution ComfyUI IMAGE tensors - a batch you can pipe into a KSampler, an upscaler, an IP-Adapter reference, or a save node. It's the bridge between "the pack found these" and "the rest of my workflow can use these."

The mechanism is unglamorous and exactly what you'd hope: it walks the result set in rank order, opens each file with Pillow, resizes anything larger than max_dimension (default 1024) down to fit, and stacks them into one image batch. Stored thumbnails are not used - this is the real file, decoded fresh. That's the point of the max_dimension cap: a batch of ten 6000px RAW renders would blow up VRAM the moment you touched it, so loading at a sensible working size and letting your upscaling stage handle the rest is the sane pattern.

Inputs

Required: results. Optional: max_images (default 4, up to 16) and max_dimension (default 1024, up to 4096). Output: images (IMAGE batch tensor).

Two practical notes. max_images defaults to just 4 because loading is expensive - don't crank it to 16 and feed a KSampler unless you're deliberately batch-processing. And like every output node here, it expects you've already filtered if you want only strong matches; max_images truncates, it doesn't curate.

Install

Same pack install as every node here - ComfyUI Manager (search "Semantic-Search") or git clone https://github.com/EricRollei/Semantic-Search into custom_nodes, install the requirements.txt deps, restart, nodes under Eric/SemanticSearch.

Where people get burned

• Files on disk are gone. If the original images were moved or deleted after indexing, this node errors or skips them - the index stores paths, not copies. That's also why you sometimes see fewer images than max_images: failed loads get dropped from the batch.
• VRAM spike when you raise max_dimension. 4096px images as tensors are big. Keep the default 1024 unless your downstream stage genuinely needs more, and raise it one notch at a time while watching the meter.
• It's for images, not clips. Result sets can contain videos and PDF pages, but Load Result Images opens image files. For a mixed result set you'll get the images and silently fewer outputs - filter result_type to images at the search node if you only want stills.
• The "why is my batch empty" moment. Empty results in → empty batch out, no error. Check the search node and any filters upstream before suspecting the loader.

Load Result Images is the node that makes search results actionable - the last step between "the pack found the exact five shots" and "those five shots are now driving my generation."