Skip to main content
AI Models

Understanding the AI Models

Nostalgia uses multiple specialized AI models plus a shared text-intelligence layer. Here's what each one does, how Photo Insight routes work, and what to expect.

How Model Selection Works

When you upload a photo, Photo Insight runs on Nostalgia's own analysis stack first. It detects damage types, color mode, face count, capture quality, and repair risk. Based on that analysis, the restore flow builds a recipe — an ordered sequence of AI tools — and selects the best model for each step automatically.

Text features follow the same rule. Captions, stories, and date estimates reuse the upload-time analysis and saved metadata when possible, so the system does not need to re-read the image on every request.

Current catalog snapshot: 35 models total — 18 active, 3 canary,10 candidate, 4 deprecated.

Restoration

Fix scratches, tears, fading, and age wear.

ModelStatusBest ForNotes
Scene Restore (default)ActiveMost photosDefault restore baseline for old family photos. Better scene-level repair than the old fake SwinIR-lite route, with room for face-specific follow-up when needed.
GFPGAN RestoreActiveGentler portrait cleanupFace-focused repair fallback. Useful when portraits need facial clean-up after or instead of scene restoration.
CodeFormer RestoreActiveDamaged portraitsPromoted from canary 2026-03-11. High-quality face restoration, particularly strong on portraits. Monitor for over-reconstruction on small or soft faces.
SwinIR RestoreCandidateBenchmark comparisonTrue SwinIR candidate for denoising and compression repair. Benchmark-only until its best task mode is validated on real family-photo scans.
FLUX KontextCandidatePremium generative evaluationPremium generative restore candidate. Benchmark-only until trust, licensing, and artifact risk are acceptable for family photos.
Topaz Dust & ScratchCandidateSurface-defect heavy photosPremium archival clean-up candidate for dust and surface defects. Expect a materially slower pass than the launch-safe default. Benchmark-only until real-photo trust scores are collected.
SUPIRCandidateHigh-detail premium evaluationSUPIR diffusion-based restoration. Best perceptual quality for high-res photos. Significantly slower and more expensive than Microsoft Old Photo — evaluate for Pro tier premium restore option only. CVPR 2024.
DiffBIRCanaryMixed blur, noise, and compressionDiffBIR blind image restoration. Handles mixed degradation (blur + noise + compression) in a single pass. Promoted to canary 2026-03-25 for heavy-damage routing. ICLR 2024.
Scene Restore (default)Active
Best forMost photos

Default restore baseline for old family photos. Better scene-level repair than the old fake SwinIR-lite route, with room for face-specific follow-up when needed.

GFPGAN RestoreActive
Best forGentler portrait cleanup

Face-focused repair fallback. Useful when portraits need facial clean-up after or instead of scene restoration.

CodeFormer RestoreActive
Best forDamaged portraits

Promoted from canary 2026-03-11. High-quality face restoration, particularly strong on portraits. Monitor for over-reconstruction on small or soft faces.

SwinIR RestoreCandidate
Best forBenchmark comparison

True SwinIR candidate for denoising and compression repair. Benchmark-only until its best task mode is validated on real family-photo scans.

FLUX KontextCandidate
Best forPremium generative evaluation

Premium generative restore candidate. Benchmark-only until trust, licensing, and artifact risk are acceptable for family photos.

Topaz Dust & ScratchCandidate
Best forSurface-defect heavy photos

Premium archival clean-up candidate for dust and surface defects. Expect a materially slower pass than the launch-safe default. Benchmark-only until real-photo trust scores are collected.

SUPIRCandidate
Best forHigh-detail premium evaluation

SUPIR diffusion-based restoration. Best perceptual quality for high-res photos. Significantly slower and more expensive than Microsoft Old Photo — evaluate for Pro tier premium restore option only. CVPR 2024.

DiffBIRCanary
Best forMixed blur, noise, and compression

DiffBIR blind image restoration. Handles mixed degradation (blur + noise + compression) in a single pass. Promoted to canary 2026-03-25 for heavy-damage routing. ICLR 2024.

Generative Reimagine

Frontier image-editing candidates for opt-in reconstruction experiments. These are benchmark-only until identity preservation and disclosure rules are proven.

ModelStatusBest ForNotes
GPT Image 2CandidateFrontier reimagine benchmarkFrontier OpenAI image-edit candidate for opt-in reimagining of damaged photos. Benchmark-only until identity preservation, disclosure copy, cost, and rate limits are acceptable.
Gemini 3.1 Flash ImageCandidateFast frontier reimagine benchmarkGoogle Nano Banana 2 / Gemini 3.1 Flash Image candidate for fast image editing and reimagining. Benchmark-only; keep outputs disclosed as generative.
Gemini 3 Pro ImageCandidateHigh-quality frontier reimagine benchmarkGoogle Nano Banana Pro / Gemini 3 Pro Image candidate for high-quality image editing. Benchmark-only; never default to factual restore without identity-safety evidence.
Imagen 4 UltraCandidateQuality reference benchmarkImagen 4 Ultra frontier comparison candidate. Include in review sheets for quality reference; not a default family-photo restore route.
FLUX.2 DevCandidateOpen-core frontier comparisonFLUX.2 frontier comparison candidate for image editing and reimagining. Keep separate from FLUX Kontext restore candidate until provider endpoint and license fit are verified.
GPT Image 2Candidate
Best forFrontier reimagine benchmark

Frontier OpenAI image-edit candidate for opt-in reimagining of damaged photos. Benchmark-only until identity preservation, disclosure copy, cost, and rate limits are acceptable.

Gemini 3.1 Flash ImageCandidate
Best forFast frontier reimagine benchmark

Google Nano Banana 2 / Gemini 3.1 Flash Image candidate for fast image editing and reimagining. Benchmark-only; keep outputs disclosed as generative.

Gemini 3 Pro ImageCandidate
Best forHigh-quality frontier reimagine benchmark

Google Nano Banana Pro / Gemini 3 Pro Image candidate for high-quality image editing. Benchmark-only; never default to factual restore without identity-safety evidence.

Imagen 4 UltraCandidate
Best forQuality reference benchmark

Imagen 4 Ultra frontier comparison candidate. Include in review sheets for quality reference; not a default family-photo restore route.

FLUX.2 DevCandidate
Best forOpen-core frontier comparison

FLUX.2 frontier comparison candidate for image editing and reimagining. Keep separate from FLUX Kontext restore candidate until provider endpoint and license fit are verified.

Face Enhancement

Refine portraits after the main restore pass when faces still need help.

ModelStatusBest ForNotes
CodeFormer Face EnhanceActiveSoft or damaged facesCodeFormer face restoration as a standalone step. Run after general restoration to enhance individual faces. Uses fidelity_weight=0.7 by default.
GFPGAN Face EnhanceActiveSubtle face cleanupGFPGAN face restoration fallback. Less aggressive than CodeFormer, useful when CodeFormer over-reconstructs soft or small faces. Improvements are often subtle on well-preserved faces — most visible when zooming in at full resolution.
OSDFaceCanaryFast one-step face enhancementOSDFace one-step diffusion face enhancement. 0.1s for 512x512 — dramatically faster than CodeFormer. CVPR 2025. Promoted to canary 2026-03-25 after benchmark review.
CodeFormer Face EnhanceActive
Best forSoft or damaged faces

CodeFormer face restoration as a standalone step. Run after general restoration to enhance individual faces. Uses fidelity_weight=0.7 by default.

GFPGAN Face EnhanceActive
Best forSubtle face cleanup

GFPGAN face restoration fallback. Less aggressive than CodeFormer, useful when CodeFormer over-reconstructs soft or small faces. Improvements are often subtle on well-preserved faces — most visible when zooming in at full resolution.

OSDFaceCanary
Best forFast one-step face enhancement

OSDFace one-step diffusion face enhancement. 0.1s for 512x512 — dramatically faster than CodeFormer. CVPR 2025. Promoted to canary 2026-03-25 after benchmark review.

Colorization

Add natural color to black-and-white photos.

ModelStatusBest ForNotes
DDColor (default)ActiveMost black-and-white photosBalanced default colorizer. Use conservatively on weak inputs; stronger scans score better. May occasionally produce muted or unnatural colors — compare before accepting and re-run if needed.
DeOldifyActiveAlternative color interpretationEconomical fallback colorizer. Useful for comparison, but generally less trustworthy than DDColor or Topaz on founder-grade photos.
Topaz (Colorize)ActiveColorizePremium colorization candidate for the quality lane. Slower than DDColor and only worth it when the still already feels trustworthy. Promote only with artifact review, not metadata alone.
InstColorCandidateSkin-tone evaluation canaryInstColor instance-aware colorization. Better skin tone handling than DDColor, supports user-guided color hints. Benchmark for skin tone accuracy across diverse photos.
DDColor (default)Active
Best forMost black-and-white photos

Balanced default colorizer. Use conservatively on weak inputs; stronger scans score better. May occasionally produce muted or unnatural colors — compare before accepting and re-run if needed.

DeOldifyActive
Best forAlternative color interpretation

Economical fallback colorizer. Useful for comparison, but generally less trustworthy than DDColor or Topaz on founder-grade photos.

Topaz (Colorize)Active
Best forColorize

Premium colorization candidate for the quality lane. Slower than DDColor and only worth it when the still already feels trustworthy. Promote only with artifact review, not metadata alone.

InstColorCandidate
Best forSkin-tone evaluation canary

InstColor instance-aware colorization. Better skin tone handling than DDColor, supports user-guided color hints. Benchmark for skin tone accuracy across diverse photos.

Denoise

Reduce scan noise and grain before heavier repair work.

ModelStatusBest ForNotes
SwinIR DenoiseActiveNoisy scansSwinIR denoising mode. Run before restoration to prevent noise amplification. Uses noise_level=25 by default.
SwinIR DenoiseActive
Best forNoisy scans

SwinIR denoising mode. Run before restoration to prevent noise amplification. Uses noise_level=25 by default.

Enhancement

Sharpen and upscale once the base repair looks trustworthy.

ModelStatusBest ForNotes
Real-ESRGANActiveLow-resolution scansReal-ESRGAN 2x super-resolution. Use after restoration, not as a substitute for restore. Does not work on grayscale images — colorize first if photo is B&W.
Real-ESRGANActive
Best forLow-resolution scans

Real-ESRGAN 2x super-resolution. Use after restoration, not as a substitute for restore. Does not work on grayscale images — colorize first if photo is B&W.

Deblur

Reduce motion blur and softness from handheld captures.

ModelStatusBest ForNotes
NAFNetActivePhone captures and photo-of-photo blurNAFNet deblurring for soft captures and photos-of-photos. Promoted to active 2026-03-18 after canary validation.
NAFNetActive
Best forPhone captures and photo-of-photo blur

NAFNet deblurring for soft captures and photos-of-photos. Promoted to active 2026-03-18 after canary validation.

Deglare

Reduce glare from glossy prints and framed photos.

ModelStatusBest ForNotes
Composite DeglareActiveMild to moderate glareLocal OpenCV glare detection (LAB bright+neutral mask) + Telea inpainting. Replaced NAFNet which was just running deblurring. Works on mild-moderate glare. Zero inference cost. Future: evaluate WindowSeat diffusion model for heavy glare.
Composite DeglareActive
Best forMild to moderate glare

Local OpenCV glare detection (LAB bright+neutral mask) + Telea inpainting. Replaced NAFNet which was just running deblurring. Works on mild-moderate glare. Zero inference cost. Future: evaluate WindowSeat diffusion model for heavy glare.

Background Removal

Cut out a subject cleanly when you need a shareable or printable silhouette.

ModelStatusBest ForNotes
rembgActiveClean subject cutoutsBackground removal via rembg.
rembgActive
Best forClean subject cutouts

Background removal via rembg.

Animation

Generate subtle motion from trusted still portraits.

ModelStatusBest ForNotes
MiniMax LiveActiveTrusted portraitsCommercial-safe still-image animation baseline. Use only after the still passes a dignity and trust check.
LivePortrait ResearchCanaryCheaper motion canaryPromoted to canary 2026-03-11. Uses system driving videos for automatic subtle motion. 20x cheaper than MiniMax at comparable portrait quality. Monitor for uncanny motion artifacts.
MiniMax LiveActive
Best forTrusted portraits

Commercial-safe still-image animation baseline. Use only after the still passes a dignity and trust check.

LivePortrait ResearchCanary
Best forCheaper motion canary

Promoted to canary 2026-03-11. Uses system driving videos for automatic subtle motion. 20x cheaper than MiniMax at comparable portrait quality. Monitor for uncanny motion artifacts.

Captions, Stories & Date Estimates

Shared text-intelligence operations reuse upload-time Photo Insight and metadata when possible, then return structured caption, story, and date outputs.

ModelStatusBest ForNotes
Claude Haiku 4.5 CaptionActiveCaptions and date estimatesClaude Haiku 4.5 caption generation.
Claude Haiku 4.5 StoryActiveLonger family narrativesClaude Haiku 4.5 family story generation.
Claude Haiku 4.5 CaptionActive
Best forCaptions and date estimates

Claude Haiku 4.5 caption generation.

Claude Haiku 4.5 StoryActive
Best forLonger family narratives

Claude Haiku 4.5 family story generation.

Known Failure Modes

AI restoration is powerful but not infallible. Here are the scenarios where models struggle or do not work, and what to do instead.

ScenarioToolSeverityAdvice
Daguerreotypes & platesRestoreDoes not workSeek professional conservation
Heavy emulsion liftingRestorePoor resultsMissing emulsion = missing data
Grayscale inputEnhanceDoes not workColorize first, then enhance
Very small facesFace EnhanceMay hallucinateCompare before/after carefully
Heavy glare (washed out)DeglareLimited resultsRescan with indirect light
Original camera blurDeblurLimited resultsRescan on flatbed if possible
Muted colorsColorizeInconsistentRe-run colorization, or try a different photo crop
Daguerreotypes & platesDoes not work
ToolRestore

Seek professional conservation

Heavy emulsion liftingPoor results
ToolRestore

Missing emulsion = missing data

Grayscale inputDoes not work
ToolEnhance

Colorize first, then enhance

Very small facesMay hallucinate
ToolFace Enhance

Compare before/after carefully

Heavy glare (washed out)Limited results
ToolDeglare

Rescan with indirect light

Original camera blurLimited results
ToolDeblur

Rescan on flatbed if possible

Muted colorsInconsistent
ToolColorize

Re-run colorization, or try a different photo crop

See the AI in action

All this technology works behind the scenes. You just upload a photo and get a result.

Understanding the AI Models · Nostalgia Family Archive