Qwen3 Text to Image

The plain sibling

There are two Qwen3 text-to-image nodes in this pack and this is the one without "Pro" in the name. Same widgets, same output sockets, same 1K/2K resolutions and eight aspect ratios - different model behind it (qwen3/text-to-image instead of qwen3/pro-text-to-image), which on a metered API means cheaper per call and, on most prompts, the difference you'd expect from a non-Pro tier: fine, occasionally great, not the last word in prompt adherence.

I'd reach for this one for the boring jobs. Contact sheets. Drafts where you're deciding whether a scene works at all. Anything you're going to reroll eight times before one image earns its place in the graph - those rerolls are exactly where the cheaper endpoint pays for itself, and where a hosted generator beats local on convenience if you don't have the card for it.

Mechanism, briefly

The node is generated from KIE's own documentation page for the endpoint, so its inputs are the documented request fields presented as widgets. Submitting posts a task to KIE with the qwen3/text-to-image model ID, the node polls until it finishes, then downloads the result. timeout_seconds defaults to 1200 and can go to 7200; a reroll that dies at the timeout still billed you.

Because the pack rebuilds the node tree from the live catalog at startup, this node's field list can grow if KIE documents a new parameter - you may see a widget appear after an update that wasn't there when you built the workflow. New widgets come with their own defaults, so old graphs keep running.

Inputs and outputs in practice

prompt is required and arrives pre-filled with a futuristic-city example, which is a nice way to sanity-check your key before you write anything. Then prompt_extend (default on - the author's own note says the model rewrites your prompt when it's enabled and that it "significantly improves results for simple descriptions"), resolution (1K or 2K), image_size (1:1 through 21:9), output_format (png or jpeg), negative_prompt, seed, nsfw_checker (default off, so filtering is your problem if you leave it off and your provider's if you don't), timeout_seconds, and callback_url - leave that empty unless you're running a server that wants webhooks.

Outputs: image, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left. Wire image onward; use task_id and raw_json when you're arguing with support about a job that came back wrong.

Install and key

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

Restart ComfyUI, then paste your KIE key once under Settings → KIE.ai Nodes Next → Connection. It's validated against KIE before replacing anything, stored outside the workflow, and shown back to you as a masked fingerprint with a ✅ CONNECTED status plus your credit balance. Servers and headless installs can skip the UI: KIE_API_KEY in the environment wins.

Honest caveats

KIE is a reseller, so you're two hops from the model - the endpoint can be repriced, re-pointed or retired, and community threads show exactly that happening to models people had built around. Keep the node's model situation in mind before you build a whole pipeline on a cheap tier. The good news is the pack's design actively expects this: it re-reads KIE's documentation catalog at most once a day and can expand your node library without a plugin update.

And the usual one for any API node: your prompt and any connected images leave your machine. For a draft sweep of forty thumbnails that's a fair trade. For client work under NDA, it isn't.