Get 4o Image Details

A lookup node for KIE's 4o image endpoint - the one that calls OpenAI's 4o image model, the thing that went viral inside ChatGPT in 2025 and that the KB still describes as excellent output wrapped in unpredictable refusals. This card is for going back to a job rather than starting one: you hand it the record of an existing task and it returns the resulting image.

When you'd actually use it

Most of the time you won't. The Generate 4o Image node in the same family waits for its own job and hands you a native IMAGE socket, so a normal generate-then-use workflow never needs to fetch anything.

This card earns its place in three situations: the generation ran in a previous session and you only kept the task record; a job that errored or timed out locally actually completed on KIE's side and you want the file without paying again; or you're polling a long job yourself instead of holding a ComfyUI run open.

What's on the card - and why it's sparse

The fields are:

prompt (required) - a generic text string.

expert_override_json - a raw JSON object merged into the request body, described by the pack as the escape hatch for "APIs whose documentation has no request schema yet."

That's it: no timeout, no callback.

The sparseness isn't neglect. The pack builds every node from KIE's published API documentation, and where a page gives response examples but no request schema, there's nothing to infer fields from, so it falls back to this generic shape. On a card like that, expert_override_json is the interface - the documented identifier for the task goes in there as JSON, named the way KIE's page for that endpoint names it. KIE also ships a sibling note under the same family about looking a job up by its ID, which is worth having open in a tab while you fill the override in.

Outputs

image - a native ComfyUI IMAGE, so it drops into a Save Image or a preview. url - the primary result URL. all_urls_json - every URL the response reported, which matters for a job lookup because a 4o image task can carry more than one artifact. task_id - the record's identifier. raw_json - the complete provider response, and the thing to read first when the image socket comes back empty: a refused or still-running job tends to say so in the body rather than raising. credits_consumed and credits_left - floats, plus the credits receipt the node draws on itself.

Install

ComfyUI Manager can find the pack by its title, or:

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

Restart. Requirements: requests, Pillow, numpy, PyYAML. No local models, no VRAM - this node is a client and each call is billed.

Key: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key; checked against KIE before it replaces a saved key, and shown afterwards as a masked fingerprint plus CONNECTED. Servers and headless setups can set KIE_API_KEY, which takes priority over the saved key.

Gotchas

JSON in a text box. The override is a string field. Malformed JSON fails locally before the request goes out - which is the cheap outcome, but it's still on you to write valid JSON.

A lookup can cost. The pack's model nodes run on metered KIE credits, so don't poll in a tight loop; find out what a re-fetch costs before you wire the node into a batch.

Old records age out. Task lookups work against KIE's records, not a local history, so an identifier from months ago is not guaranteed to resolve. Save the image and url outputs from the original run if you care about the result.

The card can gain widgets. The pack re-reads KIE's API documentation about every 24 hours, so if a request schema shows up for this endpoint later, expect real fields after a ComfyUI restart.