Kling v3 (BudgetPixel)

Most people think of Kling as a video brand, so it's easy to miss that Kuaishou also ships a serious image model - and this node is it. Kling AI v3 does text-to-image and single-reference-image generation at 1K/2K, with a strong reputation for prompt adherence and aesthetics, at a flat 35 credits per image. It even has something many of the pack's other image nodes don't: a negative_prompt input, the classic control that's quietly disappearing from hosted image APIs.

The 35-credit flat rate puts it right beside Grok Imagine on the value shelf - a "good-looking image without reading a billing matrix" price. Its differentiator over Grok is the negative prompt (real prompt-steering power) and single-reference editing. If you've been fighting a model that keeps adding the thing you didn't ask for, a negative prompt is often the fastest fix that doesn't exist on most of the pack.

How it works

Standard BudgetPixel image flow: POST your prompt, optional reference image, and negative prompt into the Kling image pipeline, poll /images/{id}, download the IMAGE batch. The single image input is a one-reference edit/composition slot - feed a subject and the prompt describes the transformation around it. It's not a multi-reference composer like FLUX 2 Dev or GPT-Image-2; it's the focused, prompt-driven workhorse.

The inputs that matter

• prompt - required, multiline.
• negative_prompt - the node's hidden gem. Tell it what to leave out ("blurry, extra fingers, watermark") and it obeys far better than most hosted models that dropped this input.
• image - optional single reference for image-to-image.
• size - 1K or 2K, flat price either way.
• aspect_ratio - framing.
• num_images - 1–4, default 1 (each at 35 credits).
• seed - default -1 (random).

Output is images (IMAGE batch).

Install

Manager (search "BudgetPixel") or:

cd ComfyUI/custom_nodes
git clone https://github.com/budgetpixel-ai/comfyui-budgetpixel.git
pip install -r comfyui-budgetpixel/requirements.txt

Restart, set BUDGETPIXEL_API_KEY or budgetpixel.json ({"api_key": "bpx_live_..."}) in the pack folder. No model files; requests is the only dependency.

Where people get burned

The trap here is category confusion - you might expect Kling-grade video from a node whose brand is Kling, but this one outputs still images only, and its video siblings are separate nodes in the BudgetPixel/Video category. Second, the flat pricing makes 2K feel free, and it mostly is (same 35 credits), but 2K renders are slower server-side - for drafts, 1K keeps iteration snappy. Third, single-reference editing is a real limit: if your composition needs multiple anchors, step up to the multi-reference models even though they cost more, because one reference won't hold a character sheet and a location at once. And use that negative prompt - it's the node's superpower and it's the first thing to reach for when the output has a recurring flaw.