Get Image Details

What it's for

Flux Kontext is the instruction-editing line from Black Forest Labs - "change the jacket to red," "put the mug on the left" - and on KIE it's resold alongside the rest of the catalog. The generation node gives you the edited image. This node gives you the record of a Kontext task: its status, its result URLs, its metadata, as JSON.

So why does it exist? Because async jobs escape graphs. You queue something, ComfyUI restarts, the workflow is re-run from a different node, or an external tool - n8n, a cron script, a colleague's dashboard run - submitted the task and you only have the ID. Get Image Details is the way back in. You hand it the task, it fetches the record.

Two things it is not: it isn't a generation node (nothing is produced, nothing is charged), and it isn't a replacement for this pack's Kontext generation node, which already waits for its own job and returns the image directly.

How it works

It calls Kontext's record-info endpoint (/api/v1/flux/kontext/record-info in the pack's bundled catalog). The pack's client already knows this API family: its internal status reader expects a successFlag on Kontext-family results, treats 1 as done, and raises on 2 or 3 with whatever errorMessage or failMsg the provider supplied. That's the useful bit - failures surface as readable messages rather than silent empties.

Now the caveat, and it's the same one that applies across this pack's "Get … Details" family. KIE's documentation page for this endpoint had no machine-readable request schema when the pack built its node catalog, so the node came out of the fallback generator. The visible inputs are therefore placeholders. The pack re-resolves unresolved pages on demand from KIE's docs, so a later UI reload can bring in the genuine documented fields - if they show up, prefer them.

Inputs and outputs

prompt is the only required input: a multiline STRING, tooltip "Prompt sent to this KIE model." For a record lookup this is a placeholder, not a parameter. Feeding it text does nothing for you.

expert_override_json (optional, default {}) is where the real work happens: "Expert-only payload overrides for APIs whose documentation has no request schema yet." On a GET node the pack sends those keys as query parameters, so {"taskId": "…"} is how you point the node at a task. Leaving it empty and filling the prompt box is the classic way to get a well-formed request about nothing.

Output sockets: image (IMAGE), url, all_urls_json, task_id, raw_json (STRING), then credits_consumed and credits_left (FLOAT). For a record read, plan on raw_json as your payload and url as the convenience link. The image socket is there because the pack derives output shape from the API family; treat it as the media kind of the record you're reading, not a guarantee of a clean tensor from a status call.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
pip install -r ComfyUI-KIE-Nodes-Next/requirements.txt

Restart, then set the key at Settings → KIE.ai Nodes Next → Connection. The pack checks the key against KIE before saving, shows only a masked fingerprint afterwards, and keeps the full key out of your workflow JSON - which matters when you share workflows. KIE_API_KEY covers headless installs and overrides the saved value.

Dependencies are requests, Pillow, numpy, PyYAML. No model downloads, no VRAM. Windows users can run the one-click installer in dist/ instead; it detects Desktop, Portable and manual installs and backs up any older copy of the pack first.

Where people get burned

The expired-reference problem. Kontext editing nodes often take an input image and the pack uploads it to KIE for you, so the inputs of a record are KIE-side URLs that have their own lifetime. If a record's URLs come back dead, that's usually expired temp storage rather than a broken node - the same reason the pack ships the URL-conversion utility nodes. Convert or download promptly.

Second, the placeholder fields again: this family reports "no data" rather than "no task ID supplied," so a blank prompt plus empty override looks like a KIE-side problem.

Third, and worth internalising for a reseller API in general: your images and prompts leave your machine on every call, the metering is per call, and the marketplace's model roster shifts. For Kontext that matters less than for video, but the habit is worth having - know what you're spending before you wire a batch loop.