Clarity Upscaler: Everything You Need to Know
Clarity Upscaler isn't just a traditional upscaler that makes pixels bigger — it uses AI to intelligently hallucinate new details as it enlarges images. The multi-stage pipeline first upscales with ESRGAN, then uses a fine-tuned Stable Diffusion model guided by ControlNet to add realistic detail that wasn't in the original image. The result is upscaled images that look like they were captured at higher resolution, not just interpolated. With controllable creativity and resemblance parameters, you can dial in exactly how much creative liberty the AI takes.
At a Glance
| Category | Upscaling |
| Creator | philz1337x |
| Released | March 15, 2024 |
| Parameters | ~1.5–2 billion across the combined pipeline (Stable Diffusion 1.5 UNet ~860M + ControlNet + ESRGAN) |
| Architecture | Multi-stage Pipeline (ESRGAN + SD 1.5 + ControlNet Tile) |
| Resolution | Up to 200x scale factor (practical limits apply) |
| License | Open source (can be self-hosted with Docker) |
| PicPresto Tier | Standard |
| Credit Cost | 3 credits per upscale |
| Approx. Cost | $0.02 per upscale |
About philz1337x
An independent AI developer based in Thailand who created Clarity Upscaler as a free, open-source alternative to Magnific AI's commercial upscaling service.
Reverse-engineered the approach used by Magnific AI (a $40/month commercial service) and released it as open source, making professional-grade AI upscaling accessible to everyone.
How It Works
A multi-stage pipeline combining: (1) 4x-UltraSharp ESRGAN for initial upscaling, (2) Juggernaut fine-tuned SD 1.5 model for diffusion-based detail enhancement, (3) ControlNet Tile for structural guidance during enhancement, and (4) MultiDiffusion tiling with 896×1152 pixel tiles for processing large images.
Training data: Component models trained on standard open-source datasets. The pipeline itself is a novel combination rather than a single trained model.
Key Innovations
- Multi-stage pipeline combining traditional upscaling (ESRGAN) with AI-generated detail enhancement (diffusion)
- Controllable 'Creativity' parameter balances faithful upscaling vs. generative detail addition
- ControlNet Tile guidance preserves structural integrity while the diffusion model adds detail
- Prompt-guided upscaling lets you describe what the image should look like for context-aware enhancement
- Open-source alternative to commercial $40/month upscaling services
Why People Love It
- Free and open-source alternative to expensive commercial upscalers like Magnific AI
- The results genuinely look like higher-resolution captures, not just bigger pixels
- Creativity control lets you decide how much the AI invents vs. preserves
- Works brilliantly on AI-generated images — the perfect companion for image generation models
- At 3 credits per upscale, it's incredibly affordable for the quality improvement
Strengths
- Dramatically improves low-resolution images with intelligently hallucinated detail
- Highly controllable: creativity, resemblance, and fractality parameters give precise control
- Open source and self-hostable with Docker
- Prompt-guided upscaling adds context-aware detail based on text descriptions
- LoRA support enables style injection during upscaling
- Excellent for enhancing AI-generated images for print or display
Limitations
- Based on Stable Diffusion 1.5 — an older base model compared to newer alternatives
- Can hallucinate incorrect details at high creativity settings
- Quality degrades at extreme scale factors
- Tiling artifacts can appear at tile boundaries if not carefully configured
- Processing time increases significantly with larger images
Best Use Cases
- Upscaling AI-generated images for print or large-format display
- Enhancing old or low-resolution photographs
- Improving product images for e-commerce
- Preparing social media content from low-res sources
- Any workflow needing increased resolution with realistic detail addition
Using Clarity Upscaler on PicPresto
Clarity Upscaler is available on PicPresto as a Standard tier model at 3 credits per upscale (approximately $0.02).
Generate an image with any model, then use the upscale button to enhance it with Clarity Upscaler.