Extensions/ComfyUI-FL-PRXPixel
ComfyUI Extension

ComfyUI-FL-PRXPixel

Photoroom PRXPixel text-to-image nodes for ComfyUI

By filliptm·Created 2 months ago·Updated 2 months ago· 0
filliptm/ComfyUI-FL-PRXPixel
Nodes2
On cloudLocal install
CategoryFL/PRXPixel
Stars0
Updated2 months ago

Nodes (2)

FL PRXPixel Text To Image
FL/PRXPixel
FL PRXPixel Model Loader
FL/PRXPixel
Readme

FL PRXPixel

Text-to-image custom nodes for ComfyUI powered by Photoroom's PRXPixel model.

Hugging Face Model PRXPixel Space ComfyUI

FL PRXPixel workflow preview

PRXPixel is a pixel-space diffusion model. It generates RGB images directly, without a VAE decode step, and this node pack exposes the model as a simple ComfyUI text-to-image workflow.

Features

  • Text-to-image generation with Photoroom/prxpixel-t2i
  • Direct ComfyUI IMAGE output without a VAE node
  • Live denoising previews on the generation node
  • Patch-count outputs for inspecting the internal denoising grid
  • Optional patch-grid controls for directly changing patch columns and rows
  • Resolution controls up to 4096x4096
  • Automatic model loading through Diffusers and Hugging Face cache

Nodes

| Node | Description | |------|-------------| | FL PRXPixel Model Loader | Loads and caches the PRXPixel Diffusers pipeline. | | FL PRXPixel Text To Image | Generates images from prompts, exposes patch metrics, and can show live diffusion previews. |

Installation

ComfyUI Manager

After this pack is available in the Comfy Registry, install ComfyUI-FL-PRXPixel from ComfyUI Manager.

Manual Install

Clone this repository into your ComfyUI custom_nodes folder:

cd custom_nodes
git clone https://github.com/filliptm/ComfyUI-FL-PRXPixel.git
cd ComfyUI-FL-PRXPixel
python -m pip install -r requirements.txt

Restart ComfyUI after installation.

Example Workflow

An example ComfyUI workflow is included at:

example workflow/PRXpixel.json

Quick Start

  1. Add FL PRXPixel Model Loader.
  2. Add FL PRXPixel Text To Image.
  3. Connect the pipeline output to the generation node.
  4. Enter a prompt and queue the workflow.

The default generation settings follow the model card baseline: 1024x1024, 28 steps, guidance scale 5.0, and bfloat16.

Model

| Model | Type | Notes | |-------|------|-------| | Photoroom/prxpixel-t2i | Pixel-space text-to-image diffusion | Generates RGB images directly. Native examples use 1024x1024, Qwen3-VL text encoding, and FlowMatch Euler scheduling. |

This node pack currently targets text-to-image only. Image-to-image, inpainting, ControlNet, and LoRA workflows are not implemented.

Key Parameters

| Parameter | Description | |-----------|-------------| | prompt | Positive text prompt. | | negative_prompt | Optional negative prompt. | | width, height | Requested output size, up to 4096x4096. | | steps | Number of denoising steps. | | guidance_scale | Classifier-free guidance strength. | | seed | Generation seed. | | use_resolution_binning | Lets the pipeline choose an internal supported resolution bin. | | use_patch_grid | Uses patch columns and rows to define the internal image size. | | patch_columns, patch_rows | Direct patch-grid controls. Each patch is currently 16x16 pixels. | | show_preview | Sends intermediate denoising frames to the ComfyUI node preview. | | preview_every | Controls how often live preview frames are emitted. |

Patch Grid

PRXPixel uses fixed-size 16x16 patches internally. When use_patch_grid is enabled, the node derives the working resolution from patch dimensions:

width = patch_columns * 16
height = patch_rows * 16
patch_count = patch_columns * patch_rows

The generation node also returns:

  • patch_count: internal patch grid size after optional resolution binning.
  • effective_patch_tokens: patch_count * batch_size * cfg_multiplier.
  • patch_info: requested/internal resolution and patch grid summary.

Verification

Standalone model smoke test:

venv\Scripts\python.exe custom_nodes\ComfyUI-FL-PRXPixel\tests\prxpixel_standalone_smoke.py

ComfyUI API end-to-end test, with ComfyUI running on port 8188:

venv\Scripts\python.exe custom_nodes\ComfyUI-FL-PRXPixel\tests\prxpixel_comfy_api_e2e.py

Requirements

  • Python >=3.10
  • ComfyUI
  • CUDA GPU with bf16 support strongly recommended
  • Hugging Face model cache space for the PRXPixel weights, roughly 17.5 GB

The current PRXPixel model card references a Diffusers development branch. This pack installs the matching public Space wheel:

https://huggingface.co/spaces/Photoroom/PRX-Pixel/resolve/main/diffusers-0.39.0.dev0-py3-none-any.whl

License

Apache-2.0