P-Image (BudgetPixel)

P-Image is Pruna AI's optimized Qwen-based image model, and it is the cheap-shot node of this pack. Ten credits per image, about a second per render, one megapixel output. That's the entire pitch, and it's a good one: this is the node you point at the wall of concept variations, the fifty thumbnails you don't care about individually, the "generate ten takes of this and I'll pick one" workflow. It won't beat Nano Banana on polish and it can't do the agentic stuff Muse does, but at 10 credits it doesn't need to - it needs to be fast and nearly free, and it is.

Like every node in the BudgetPixel pack it's a hosted API call. Prompt goes up, the optimized Qwen pipeline runs server-side in about a second, and the image comes back as a core IMAGE tensor. The model also auto-enhances prompts - the tooltip says so - which means you can feed it sloppy text and still get a coherent result. That's a feature for high-volume work: less time writing the perfect prompt, more time looking at outputs.

The inputs

• prompt (required) - text description. The model enhances it automatically, so don't over-engineer.
• seed - default -1 (random), or set it for reproducible generation. This is the control that makes batch iteration useful: lock a seed, vary the prompt, compare.
• aspect_ratio - seven options, 1:1 default.
• num_images - 1… and that's it. max is 1. This model returns a single image per request. If you want ten, run the node ten times.

Output is images into Preview Image or Save Image.

The tradeoff, honestly

P-Image is a Qwen-derived model at a fraction of the cost of the pack's dedicated Qwen nodes (15–85 credits). The quality gap is real but the use case is different: Qwen Image is for the image you care about; P-Image is for the image you need a lot of. It's also genuinely fast - "about a second" is the model's whole positioning, and it holds up. For iteration loops, that's the difference between a workflow that flows and one that stalls waiting on a render.

One honest caveat: prompt auto-enhancement is a double-edged sword. It makes bad prompts survivable, but it also means your carefully worded negative constraints can get washed out. For precise control, spend the extra credits on a Qwen Image node; for volume, P-Image.

Install and issues

Install via ComfyUI 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, find it under Add Node → BudgetPixel → Image, and set your API key via BUDGETPIXEL_API_KEY or a budgetpixel.json file - no key widget, by design.

Pack-wide caveats: a recent ComfyUI for core types, and interrupting a run doesn't cancel the server-side job - though at 10 credits and ~1 second, the interruption trap is barely worth worrying about here, which is part of the appeal.