Higgsfield - Result Images

The two nodes that end in a fenced-off result - Custom Model (Advanced) and Resume Request - don't hand you an image. They hand you an HF_RESULT, a bundle of downloaded file paths and request metadata. That type only exists inside this pack, and nothing else in ComfyUI knows what to do with it. HFResultImages is the translator: HF_RESULT in, a normal IMAGE list out, plus the paths on disk and the request ID.

Put it this way - this is the node that lets a closed model's output rejoin your local pipeline. Upscale it, crop it, mask it, run it through IC-Light, feed it back in as a reference for the next generation. From the moment it leaves this node it's just another image tensor, and the rest of your workflow has no idea where it came from.

How it works

Almost nothing, which is the point. When the generating node finishes, the media has already been downloaded into ComfyUI/output/higgsfield/<your-key-folder>/ - so this node does no API call and no downloading, it just reads those files back off disk. Each file is opened, EXIF-rotated if needed, converted to RGB, and pushed through as a 0–1 float tensor. It also registers each as a preview, so the picture shows up in the node itself when the graph runs.

The one validation it does is the useful kind: point it at a request that produced no images and you get "This request did not produce images" rather than a confusing empty output. That's your cue that you want Result Video instead.

Inputs and outputs

result takes the HF_RESULT from Advanced Generate or Resume Request. That's the whole input list.

Out come three things. images is an IMAGE list - one entry per returned image, in the order the API returned them - and because it's a list, a downstream single-image node runs once per image. Wiring a Sunburst request that produced four images into a Save Image node gives you four saved files, not a batch your saver chokes on. saved_paths is a JSON string containing the on-disk locations, which is handy for a text node or a script that wants to move them somewhere else. request_id is the same ID you'd paste into Resume Request.

The node is flagged as an output node, so it previews and it works fine as the last node in a workflow if all you wanted was to look at the result.

Install

Nothing extra; the pack installs as one unit:

cd ComfyUI/custom_nodes
git clone https://github.com/w0ver/Higgsfield-api-comfyui-nodes ComfyUI-Higgsfield

Restart and it's under Higgsfield/Advanced, or find it by searching "Higgsfield" in ComfyUI Manager. No weights, no extra wheels - the pack's only requirement is requests, and credentials (the Windows Configure API.bat helper, or HF_API_KEY_ID / HF_API_KEY_SECRET in the backend environment) matter only for the nodes that actually talk to the API.

Where people get burned

• It errors on a video request. Correct behaviour - use Result Video. Mixing them up is the single most common thing people do wrong with the advanced path.
• You deleted the output folder. This node reads files, so if ComfyUI/output/higgsfield/... is gone, so is your result. Recoverable without new charges if the request is still in the local database: queue Resume Request with that request_id and it downloads again.
• The images list trips a node that expects exactly one image. Some downstream nodes run per-image, some want a batch. If you need a batch, batch them deliberately rather than hoping.
• Expecting this node to fire off a generation. It never does. It's pure local file I/O, so queueing it costs nothing - the money was spent one node upstream.