P-Image-Ideogram

P-Image is Pruna AI's image model, and Pruna's whole identity is speed. The company built its name on making open models run faster - its "FLUX-Juiced" release pushed 2.6x faster FLUX.1 [dev] - and when P-Image landed, the community reaction thread led with the one number that mattered: "It's incredibly fast, 4 image in 6 seconds." This node serves that model through Runware's cloud, which means the speed pitch survives into ComfyUI without you installing anything Pruna-specific. It's a closed, API-only model, so this hosted node is the on-ramp.

Be clear-eyed about the family resemblance before you expect too much. Early community testing pegged P-Image's output quality as similar to Z-Image - Alibaba's fast 6B model - and noted its one notable quirk: JSON-style structured prompting "doesn't work at all... it'll start writing nonsense into the image." If you were hoping for layout-precise prompting, this isn't that model. What P-Image is for is volume and iteration: fast, good-enough images cheaply, the kind of thing you run hundreds of times while exploring an idea. Its lineage shows in the node's settings - settings.thinkingLevel (default high) controls internal reasoning depth before rendering, and settings.enhancePrompt (default on) rewrites your prompt for better results, so it's an LLM-guided generator rather than a raw diffusion box.

Inputs are the familiar modern set. positivePrompt is required, sentence-style. referenceImages supplies a reference, and the advancedFeatures.watermark.* group handles stamped watermarks if you need them. width and height default to 1024, seed and numberResults behave normally, and safety.checkContent adds a moderation pass when you need it. outputFormat defaults to JPG with outputQuality alongside. Output is image.

Install

Standard Runware pack setup: ComfyUI Manager → search Runware → install → restart, or clone https://github.com/Runware/ComfyUI-Runware into custom_nodes and pip install -r requirements.txt. Then a Runware API key from the dashboard - Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

Who this is for

If your workflow is "generate a lot, keep the rare good one," P-Image-Ideogram is worth benchmarking against your local fast model - and that's the honest test, because Z-Image-class quality runs locally on modest cards for the cost of electricity, so the hosted version has to win on convenience or batch throughput. If you need structured layout control, skip it (no JSON prompting, per the community reports). If you need speed and don't want to babysit a local install, this is the cloud answer. Per-call cost applies, your prompts transit Runware's cloud, and - as with the rest of this pack - there's no open-weight fallback because P-Image is closed by design.