AI Photo Restoration Workflow 2026

The Modern Restoration Stack

Before AI, photo restoration was entirely manual — a skilled retoucher spending 2–8 hours per photo. Today, AI handles the time-consuming work of resolution enhancement and texture reconstruction, leaving manual effort focused on damage that AI can't fix: physical tears, chemical stains, missing sections.

The result: a workflow that takes 15–45 minutes per photo instead of 4–8 hours, while producing output indistinguishable from traditional restoration for most clients.

Phase 1: Digitization

The quality ceiling is set here. A mediocre scan cannot be saved by AI. Use a flatbed scanner at 600–1200 DPI. For slides and negatives, use a dedicated film scanner or a DSLR copy stand setup.

Save the raw scan as TIFF (lossless). Never apply sharpening or noise reduction at the scan stage — that's for later. Capture the damage faithfully; you'll decide what to fix in post.

Phase 2: Dust and Scratch Removal

Before AI upscaling, remove dust, scratches, and minor stains using Photoshop's Healing Brush or Content-Aware Fill. Work at 100% zoom. This is non-negotiable: AI will faithfully enlarge any artifact you leave in.

For chemical stains and color shifts, apply Curves or Selective Color adjustments to neutralize cast. The goal is a "clean" version of the degraded photo — not necessarily vibrant or corrected for final output, just damage-free.

Phase 3: AI Upscaling

Export your pre-processed scan as JPEG (95 quality) or PNG and upload to UpRes. Model selection matters here:

• Portraits: Visage restores faces and leaves the rest of the frame alone. Atelier adds denoise and a print upscale when the photo looks like a portrait.
• Landscapes: Flare — fast everyday upscale with sharp edges and nothing invented.
• Mixed scenes: Lumen — recoverable detail for print, up to 8×.
• Fine art: Lumen when the brushwork was really there. Mirage only when invented detail is acceptable.

Target up to 8K for standard output. Use 8K if the client will print larger than 20 inches or needs archival-grade resolution.

Phase 4: Final Correction and Output

Import the AI-upscaled output back into Lightroom or Photoshop. Apply final color correction, tonal adjustments, and optional sharpening. At up to 8K, fine-tuning is easier and less destructive.

For archival delivery, export as TIFF or PNG at full resolution. For client delivery, export as JPEG (90–95 quality). For large-format printing, use the full-resolution PNG without additional sharpening — the lab's RIP software handles that.

Automating the Workflow with the UpRes API

Restoration studios processing high volumes can automate the AI upscaling step via the UpRes API. Submit a photo URL, receive a webhook when processing completes, and integrate directly into your asset management system. The Business plan includes API access and priority processing queue.

See the batch upscaling API guide for a complete integration walkthrough.