Zhenzhen_Flux_2_Flex

Flux 2 is Black Forest Labs' current image model line, and the Comfly pack wraps the three API tiers - Flex, Max, Pro - as separate nodes. Comfly_Flux_2_Flex is the "tuned" one: it's the only of the three that exposes the sampling dials (guidance, steps, prompt_upsampling), which makes it the node you'd actually iterate with. The catch, as always with this pack: nothing runs locally. You're renting Flux 2 Flex by the image through the Comfly relay, so it's a paid call every time.

Inputs that matter

• prompt - required, multiline. The one thing you must set.
• input_image … input_image_8 - up to eight reference images. Feed one and the prompt becomes an edit/reimagination; several lets you blend characters or styles. This is the same "image comes along for the ride" design the whole Flux 2 line shares.
• prompt_upsampling - default on. Flex is built for this: it expands your short prompt into a richer one before generating. Leave it on unless you've hand-tuned a long prompt and don't want it touched.
• guidance - 1.5–10, default 5. How hard the generation clamps to your (upsampled) prompt. 5 is the BFL default; lower for looser, higher for literal.
• steps - 1–50, default 50. This one actually matters on Flex; fewer steps are cheaper/faster but you lose quality fast below ~25.
• width / height - 64–6000, step 8, default 1024 square.
• safety_tolerance - 0–5, default 2. Higher = stricter content filtering. The dial exists because it's an API model, and it's a paid call - don't fight it, just know it's there.
• output_format - png (default) or jpeg.

Outputs: image (IMAGE), image_url, and response.

Install

Same as every Comfly node:

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

Or ComfyUI Manager → "Comfyui_Comfly" → restart. No model downloads - it's an HTTP client. Key from https://ai.comfly.chat in Comflyapi.json (or the node's apikey).

Gotchas

The three Flux nodes look identical and that's the trap: Max and Pro have no guidance or steps, because those tiers don't accept them - so if you load a Flex workflow, swap in Max, and wonder why your settings vanished, that's by design. Flex is the tier where your sliders do something; if you're not using them, you're paying Flex prices for a Max-style experience and might as well use Max. And as a paid relay: check response when output looks off - the task log at ai.comfly.chat tells you what the API actually returned.