Flux Klein T2V Prompt Generator 🐑

Same idea as the pack's other prompt generators, pointed at Flux Klein: you feed it a rough sentence and an LLM connection, and it rewrites your idea into a detailed, model-shaped prompt. It's the "I don't want to hand-craft a paragraph every time" node, tuned for a Flux Klein target.

It's part of ComfyUI-MieNodes (ComfyUI_MieNodes), MieMieeeee's utility pack, under 🐑 Prompt Generator.

What "Flux Klein" is

Worth a quick grounding, because the name is new. Per the KB's Flux 2 panel, Klein is Black Forest Labs' compact tier of Flux 2, released January 2026 as size-distilled 4B and 9B checkpoints. It was BFL's answer to the complaint that full Flux 2 needed 18–24GB of VRAM - Klein 4B fits in about 13GB, is Apache 2.0, and generates in under a second distilled, which put a genuinely usable Flux 2 back on consumer cards. Klein 9B is the stronger, non-commercial tier and became one of Z-Image's main rivals for editing.

The node's T2V label reads oddly against that (Klein is BFL's image line), so don't over-read it - treat this as a prompt generator that produces text tuned for the author's Flux Klein pipeline. What the node does is unambiguous and that's what matters: it turns your idea into a fuller prompt via an LLM. Where you send that prompt is your graph's business.

How it works

Three inputs, one string out - the same shape as the Hunyuan T2V generator. It takes an LLM connector, your input_text, and a seed, and asks the language model, steered by a Klein-specific system prompt, to expand your rough idea into a detailed prompt. The prompt-engineering KB is clear that this system-level framing is what actually steers the LLM, so the value here is that MieMieeeee has pre-written that framing for you. The seed makes it reproducible and doubles as a reroll: change it to get a different take on the same idea.

The usual gotcha applies - the LLM itself comes from a separate connector node. MieNodes ships a family of Set…LLMServiceConnector nodes (several providers, some with free tiers); you configure one with an API key and wire it into this node.

The inputs and outputs that matter

• llm_service_connector (LLMServiceConnector) - required, from a Set-connector node. This is the piece that does the actual rewriting.
• input_text (STRING, multiline) - your rough prompt. One line is enough.
• seed (INT) - reproducibility and reroll.

Output is flux_klein_t2v_prompt (STRING). Wire it into your text encode / positive conditioning.

Installing it

ComfyUI Manager → search ComfyUI-MieNodes → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/MieMieeeee/ComfyUI-MieNodes

then restart. The node is light; using it means an LLM call, so you need a connector node with a valid API key. Nodes appear under 🐑 MieNodes.

Common issues & troubleshooting

No output. The llm_service_connector is required and comes from a separate node you have to add and configure with an API key. That missing wire is the usual culprit.

API/provider errors come from the connector - check the key, remaining credits, and the model id there, not on this node.

The prompt style doesn't suit your model. This generator is tuned for a Klein target. If you're actually generating on a different model, the phrasing may not be ideal - either accept it as a strong starting draft you edit, or use the generator that matches your model. The output is plain text, so nothing stops you from tweaking it before it hits the encoder.

Same result each run. That's the fixed seed doing its job. Change it to reroll.