P-Image-Edit

P-Image-Edit is the editing sibling of Pruna's P-Image in the Runware pack, and its whole personality is in the required inputs: instead of a blank canvas, you hand it referenceImages and tell it what to change. It's an image-to-image edit model - swap the background, change the outfit, restyle a product shot - that runs on Runware's cloud and returns a native IMAGE. Where a local img2img pipeline means fiddling with denoise strength and ControlNet, this node just takes a reference and an instruction.

That's the pitch, and it's a real one for a beginner: one reference image, one sentence, no strength slider to misjudge. The model handles the "how much to change" internally.

What you actually set

• referenceImages - required, an IMAGE socket. The image you're editing. Feed it from a Load Image node or anything upstream.
• positivePrompt - required. State the edit as a change: "replace the sky with a starry night," "turn the dress red," "remove the coffee cup."
• width / height - optional here, default 1024×1024.
• providerSettings.prunaai.turbo - turbo mode for faster generation. The tooltip is honest: it may reduce subtle details on complex tasks. Off by default.
• numberResults - up to 4 variations of the edit.
• outputFormat - JPG, PNG, WEBP.
• safety / safety.checkContent - content checking, off by default.

Output: one image socket, type IMAGE.

Install

Standard pack install: ComfyUI Manager → search Runware → install → restart, or

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

API key into ComfyUI Settings → Runware API key or RUNWARE_API_KEY. Light deps, no model downloads.

Where people get burned

• The reference sets the ceiling. If the source image is low-res or has baked-in artifacts, the edit inherits them. Start from a clean frame.
• "Edit" doesn't mean "preserve." Without a mask or a more specific instruction, the model can interpret the whole image as fair game. Be explicit about what stays.
• Turbo is tempting and it shows. On a simple edit, turbo is fine; on a complex scene, the lost detail will annoy you. Test once before you default to it.
• Per-run cost. Same as every node here - the title bar prints it, and batches multiply it.

The takeaway: for quick, instruction-driven edits without a local img2img rig, this is the node. It's not a precision tool - that's what masks and ControlNet are for - but for "change this thing, keep the rest," it's remarkably painless.