CoolFace
Apppublic

lilcats/pixelpaw

sourceHugging Faceupdated 3mo agoView on Hugging Face
0likes
App README

PixelPaw

PixelPaw is a Gradio text-to-image workbench deployed as lilcats/pixelpaw on Hugging Face Spaces ZeroGPU.

  • —Live app: https://lilcats-pixelpaw.hf.space
  • —Space repo: https://huggingface.co/spaces/lilcats/pixelpaw
  • —Hardware: zero-a10g
  • —SDK: Gradio

Current Features

  • —Lazy model selector with 12 public Diffusers text-to-image models.
  • —ZeroGPU-wrapped generation via @spaces.GPU.
  • —Prompt, negative prompt, dimensions, steps, guidance, seed, scheduler, and batch controls.
  • —Custom HTML header, SVG favicon, responsive CSS, and animated generation status panel.
  • —JSON run metadata for model, scheduler, seed, size, batch, elapsed time, and device.

Model Set

General models:

  • —stabilityai/sdxl-turbo
  • —stabilityai/sd-turbo
  • —stabilityai/stable-diffusion-xl-base-1.0
  • —playgroundai/playground-v2.5-1024px-aesthetic
  • —RunDiffusion/Juggernaut-XL-v9
  • —SG161222/RealVisXL_V5.0
  • —Lykon/dreamshaper-xl-1-0
  • —Lykon/dreamshaper-8
  • —SG161222/Realistic_Vision_V6.0_B1_noVAE
  • —cagliostrolab/animagine-xl-4.0

Adult-capable models:

  • —Heartsync/NSFW-Uncensored
  • —UnfilteredAI/NSFW-gen-v2

Preloaded models:

  • —stabilityai/sdxl-turbo
  • —stabilityai/sd-turbo

Other models are loaded lazily on first use with the Space HF_TOKEN secret, which avoids anonymous Hub rate limits. Large cold SDXL models can still be slow the first time because the Space has to download and instantiate several GB of weights.

Configuration

Environment variables:

  • —PIXELPAW_DEFAULT_MODEL: model label selected at startup. Defaults to SDXL Turbo.
  • —PIXELPAW_MODEL_ID: optional custom Diffusers model ID appended to the selector.
  • —PIXELPAW_MODEL_VARIANT: variant used for the optional custom model. Defaults to fp16.
  • —PIXELPAW_ZERO_GPU_SIZE: ZeroGPU size for decorated generation calls. Defaults to large.
  • —PIXELPAW_CACHE_DIR: writable runtime cache root. Defaults to /tmp/pixelpaw-cache.
  • —PIXELPAW_LOCAL_CPU=1: force CPU loading for local smoke tests.
  • —HF_TOKEN: Space secret used for authenticated Hugging Face Hub model downloads.

At startup PixelPaw redirects Hugging Face Hub, Xet, Diffusers, Transformers, Torch, pip, and Gradio temp caches to PIXELPAW_CACHE_DIR, disables Hub Xet by default, disables Gradio SSR, and suppresses the benign Python event-loop invalid-file-descriptor cleanup warning seen in the Space runtime.

Interface Assets

  • —assets/header.html: custom Gradio header markup.
  • —assets/pixelpaw.css: responsive styling, mobile layout, status panel, and loading animation.
  • —assets/pixelpaw-icon.svg: app icon and favicon.

Deploy

Create or update the Space:

bash
hf repos create lilcats/pixelpaw --type space --space-sdk gradio --flavor zero-a10g --exist-ok
hf upload lilcats/pixelpaw . --type space --exclude .git/** --exclude .codex/** --exclude .agents/** --exclude __pycache__/**
hf spaces wait lilcats/pixelpaw --timeout 20m