Image Super-Resolution - Fixed & Working (AI Upscaling)
# Image Super-Resolution - Fixed & Working (AI Upscaling Library)Upscale Images 4X With AI. No Monthly Fees. Works Offline.The most popular AI image upscaling library on GitHub (4,800+ stars) broke in 2024. This is the fixed version that actually works.---## The ProblemYou need to upscale images. You found the perfect AI library. You tried to install it.💥 Installation failed.```ERROR: Could not find a version that satisfies the requirement h5py==2.10.0pip install failed```You tried:- Different Python versions- Virtual environments- Stack Overflow solutions- GitHub issues (107 open, no fixes)Nothing worked. The project was abandoned.So you either:- Paid for online AI upscaling ($10-30/month)- Used lower-quality tools- Gave up on AI upscaling---## The SolutionWe fixed it.Diagnosed the dependency hell. Updated to modern Python. Tested on Colab, Kaggle, and local environments.It works now.One command. Upscale any image 4X with AI. Run locally. No monthly fees.---## What You Get✅ Working AI upscaling library - Installs successfully, runs perfectly✅ Pre-trained models - Ready to use, no training required✅ 4X upscaling - 512px → 2048px with AI enhancement✅ Multiple models - RDN, RRDN optimized for different use cases✅ GPU & CPU support - Works with or without NVIDIA GPU✅ Colab/Kaggle compatible - Works in notebooks✅ Full documentation - What was broken, how we fixed it, how to use it✅ GitHub repo access - Fork it, modify it, use commercially---## What It DoesAI Image Super-Resolution:- Upscale low-res images 2X or 4X- Enhance details with AI (not just interpolation)- Restore old/degraded photos- Prepare images for print- Video frame enhancement- AI art upscalingPowered by:- Residual Dense Networks (RDN)- Residual in Residual Dense Network (RRDN)- Pre-trained on ImageNet- PSNR and GAN-based models---## Who This Is For✅ Photographers enhancing old photos✅ Print shops needing high-res from low-res✅ AI artists upscaling generated images✅ Researchers running experiments✅ Developers building image enhancement apps✅ Anyone who needs offline AI upscaling✅ People tired of $10-30/month subscriptions---## Requirements- Python 3.8+ (works on 3.8, 3.9, 3.10, 3.11)- 4GB+ RAM (8GB+ recommended)- Optional: NVIDIA GPU with CUDA (faster, but works on CPU)- Works on: Mac, Windows, Linux, Colab, Kaggle---## Why This Works When the Original Doesn'tThe Problem: The original used 5-year-old dependencies incompatible with modern Python:- h5py 2.10.0 (2019) - Broken on modern NumPy- tensorflow 2.* - Too vague, pulled incompatible versions- Python 3.6 requirement - End of life since 2021The Fix: Updated to modern, stable versions:- h5py >=3.1.0 - Compatible with modern systems- tensorflow >=2.10,<2.18 - Stable, tested range- numpy <2.0 - TensorFlow compatibility- Python 3.8+ - Modern, maintainedThe Result: Works on Colab, Kaggle, modern Python, everywhere.---## Usage Example```python# Install (one command)pip install git+https://github.com/dmorsepgh/image-super-resolution.git# Use (three lines)from ISR.models import RDNmodel = RDN(weights='psnr-large')upscaled_img = model.predict(your_low_res_image)```That's it. Your image is now 4X larger with AI-enhanced details.---## Compare Your Options| Option | Cost | Quality | Privacy | Speed ||--------|------|---------|---------|-------|| Online AI upscaler (Topaz, etc.) | $10-30/month | High | Uploads to cloud | Fast || Free online tools | Free | Low-Medium | Public upload | Slow || Original broken library | Free | N/A | N/A | Doesn't work || This fixed version | $27 once | High | Local only | GPU: Fast / CPU: Medium |Your images stay on your machine. No uploads. No monthly fees.---## Pricing$27 - One-time paymentWhat you save:- $10-30/month in subscription fees (saves $120-360/year)- Hours debugging the broken original- Privacy (no cloud uploads)ROI: Pays for itself in 1 month vs subscriptions.---## What's Included📦 Fixed codebase - Works on modern Python📦 Pre-trained models - RDN, RRDN ready to use📦 Documentation - Installation, usage, examples📦 Technical breakdown - What was broken, how we fixed it📦 GitHub repo access - Fork, modify, use commercially📦 Free updates - Any future fixes included---## GuaranteeIf pip install doesn't work, full refund. No questions asked.We tested this on Python 3.8-3.11, Colab, Kaggle, Mac, and Linux. It works.---## Use Cases### Photography- Restore old family photos- Enhance low-res scans- Prepare images for large prints### AI Art- Upscale Stable Diffusion/Midjourney outputs- Enhance details in generated images- Create high-res versions for print### Video- Upscale video frames- Enhance old footage- Improve streaming quality### Development- Build image enhancement apps- Integrate into workflows- Research and experiments---## FAQQ: Is this legal?A: Yes. The original is open source (Apache 2.0). This is a fixed fork with proper attribution.Q: Do I need a GPU?A: No. Works on CPU (slower) or GPU (faster). Any modern computer works.Q: How does this compare to Topaz AI?A: Similar quality, but offline and no subscription. Topaz has more features, this has no monthly fee.Q: Will this work on my Mac M1/M2?A: Yes. CPU mode works great on Apple Silicon.Q: What about GIMP/Photoshop upscaling?A: Those use interpolation. This uses AI trained on millions of images. Much better quality.Q: Can I use this commercially?A: Yes. Use it for client work, products, whatever you want.Q: What if I get stuck?A: Email support included. I'll help you get it working.---## About MeI'm Doug Morse. I find valuable abandoned open source projects, fix them, and redistribute them.4,815 people starred this AI upscaling library because it was one of the best. Then it broke. I fixed it.---## Get Started Now$27 - One-time paymentDownload immediately after purchase:- Complete fixed codebase- Pre-trained AI models- Full documentation- GitHub access- Email supportStop paying monthly. Start upscaling locally.[Buy Now]---## Fine Print- One-time payment, yours forever- No subscription, no recurring fees- 30-day money-back guarantee- Use personally or commercially- Free updates included- Support: doug@dmpgh.com---Questions? Email: doug@dmpgh.comLicense: Apache 2.0 (same as original)
Get it → dougster721.gumroad.com