AxForge
Sign in Get started

Model reference · open weights

onnx-image-models

Available as managed deployment Image huggingworld · community Image edit 1 variants 3k dl/mo

onnx-image-models is an open-weight image model from huggingworld. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Request deployment All image models

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released byhuggingworld
TypeImage models
TaskImage edit
Runs withml-agents
Released2026-03-17
Popularity3k downloads / month
LicenceOpen weights

About

What onnx-image-models is

TIER 1: TOP PERFORMERS ⭐⭐⭐

  1. 4x UltraSharp V2 FP32 (51.8MB) - The gold standard for general-purpose upscaling

    • Superior detail preservation across all content types
    • Excellent handling of both photos and artwork
    • Minimal artifacts, natural texture enhancement

  2. 4x PurePhoto RealPLSKR (29.9MB) - Best for photographic content

Read the full model card
  • Exceptional on portraits and realistic images
  • Advanced noise handling while preserving skin textures
  • Recommended by multiple 2026 benchmarks

  • 4x ClearReality V1 FP32 (1.7MB) - Best lightweight option

    • Surprisingly excellent quality for its tiny size
    • Fast processing with minimal resource usage
    • Great balance of quality vs. performance

    • TIER 2: EXCELLENT SPECIALISTS ⭐⭐

    1. 4x Nomos8k DAT (89.9MB) - JPEG compression specialist

      • Specifically trained for compressed image restoration
      • Excellent on heavily compressed photos
      • Recommended for social media/web images

    2. 4x PurePhoto SPAN (1.7MB) - Another excellent lightweight

      • Very similar quality to ClearReality
      • Great for mobile/edge deployment
      • Recommended for real-time applications

    3. 4x LSDIR (67.2MB) - Large dataset trained

      • Good general performance across content types
      • Solid choice for batch processing

    TIER 3: SOLID PERFORMERS ⭐

    1. 4x UltraSharp FP32 (67MB) - Original UltraSharp

      • Still excellent but superseded by V2
      • Good for legacy compatibility

    2. RealESR-General WDN x4v3 (4.9MB) - Compact and efficient

      • Recommended in multiple 2026 reviews
      • Good balance of speed and quality

    3. 4x Nomos2 HQ DAT2 FP32 (52.9MB) - High-quality alternative

      • Similar performance to top tier but larger file size

    TIER 4: SPECIALIZED/NICHE

    1. 4x Upscayl UltraSharp (67.2MB) - Upscayl version

      • Good performance but redundant with standalone UltraSharp

    2. 4x Upscayl High Fidelity (67.2MB) - Upscayl variant

      • Quality varies by source image

    3. 2x Real-ESRGAN x2plus (67.2MB) - Only 2x scaling

      • Limited by scale factor but good quality

    TIER 5: AVOID/OUTDATED

    13-42. Most remaining models fall into these categories:

    • Outdated architectures (original ESRGAN variants)
    • Overly specialized (JPG restoration models for specific compression levels)
    • Redundant (multiple similar Upscayl variants)
    • Limited scale factors (1x models for restoration only)

    KEY RECOMMENDATIONS:

    For General Use: Start with 4x UltraSharp V2 FP32 - it consistently ranks #1 in 2026 benchmarks

    For Mobile/Limited Resources: Use 4x ClearReality V1 FP32 or 4x PurePhoto SPAN - tiny files, excellent quality

    For Compressed Images: 4x Nomos8k DAT handles JPEG artifacts better than anything else

    For Photography: 4x PurePhoto RealPLSKR is specifically optimized for photos and portraits

    The field has matured significantly, with newer architectures like DAT2 and RealPLKSR clearly outperforming older ESRGAN-based models. File size doesn't always correlate with quality - some of the best models are surprisingly compact.

    From the published model card. Full card on the HuggingFace links in the sidebar.

    How it works

    How image models work

    Text promptwhat to makeText encoderunderstands itDiffusion stepsdenoise to pixelsImagePNG / JPEGA diffusion model starts from noise and denoises it, guided by your prompt, into a finished image.

    Using it via the API

    Call it like any OpenAI endpoint

    Once AxForge deploys onnx-image-models for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (onnx-image-models below is illustrative; you get the exact model name on deployment.)

    $ curl -sS https://api.axforge.ai/v1/images/generations \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"onnx-image-models","prompt":"a red bicycle","size":"1024x1024"}'

    Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

    Get API access
    © 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms