AI Restoration of Damaged Photos: Pipeline and Deployment

Scanned film from family archives often contains scratches, abrasions, fading, and pixel artifacts. Manual restoration of a single frame in Photoshop takes 2–3 hours. Even an experienced retoucher spends up to 8 hours on complex 19th-century portraits. We automate this process through a combination of specialized models — achieving results in seconds without loss of quality. The cost of processing one image in the pipeline can be tens of times lower than manual labor, especially for mass archive restoration. Budget savings when switching from manual processing reach 80%.

Our pipeline solves the full range of issues: scratch and stain removal (inpainting), face restoration (GFPGAN), upscaling with detail enhancement (Real-ESRGAN), denoising, and color correction. All of this is combined into a single service with REST API or gRPC.

How We Restore Photos: Step by Step

1. Analyze the original image — determine damage type, resolution, presence of faces.
2. Remove scratches and stains — apply inpainting model (LaMa or internal UNet-based).
3. Upscaling — Real-ESRGAN increases resolution by 4x while preserving textures.
4. Face restoration — GFPGAN corrects features using StyleGAN2 as a prior.
5. Final post-processing — denoising, color correction, and export.

The pipeline is implemented in Python using PyTorch and ONNX Runtime. Example code below.

from PIL import Image
import cv2
import numpy as np
import io

class PhotoRestorationPipeline:
    def restore(self, damaged_photo: bytes) -> bytes:
        # 1. Scratch and stain removal (GFPGAN + inpainting)
        # 2. Upscaling (Real-ESRGAN)
        # 3. Face restoration (GFPGAN)
        # 4. Denoising

        image = Image.open(io.BytesIO(damaged_photo)).convert("RGB")
        img_np = np.array(image)

        img_np = self.remove_scratches(img_np)
        img_np = self.upscale(img_np)
        img_np = self.restore_faces(img_np)

        result = Image.fromarray(img_np)
        buf = io.BytesIO()
        result.save(buf, format="PNG")
        return buf.getvalue()

Upscaling Implementation (Real-ESRGAN)

from basicsr.archs.rrdbnet_arch import RRDBNet
from realesrgan import RealESRGANer

model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
upsampler = RealESRGANer(
    scale=4,
    model_path="RealESRGAN_x4plus.pth",
    model=model,
    tile=512,
    tile_pad=10,
    pre_pad=0,
    half=True
)

def upscale_image(img_np: np.ndarray, scale: int = 4) -> np.ndarray:
    output, _ = upsampler.enhance(img_np, outscale=scale)
    return output

Face Restoration (GFPGAN)

from gfpgan import GFPGANer

gfpgan = GFPGANer(
    model_path="GFPGANv1.4.pth",
    upscale=2,
    arch="clean",
    channel_multiplier=2
)

def restore_faces(img_np: np.ndarray) -> np.ndarray:
    _, _, restored_img = gfpgan.enhance(
        img_np,
        has_aligned=False,
        only_center_face=False,
        paste_back=True,
        weight=0.5
    )
    return restored_img

REST API with FastAPI

from fastapi import FastAPI, UploadFile, File

app = FastAPI()
pipeline = PhotoRestorationPipeline()

@app.post("/restore")
async def restore_photo(file: UploadFile = File(...)):
    original_bytes = await file.read()
    restored_bytes = pipeline.restore(original_bytes)
    return Response(content=restored_bytes, media_type="image/png")

Why GFPGAN Handles Faces Better?

GFPGAN uses StyleGAN2 as a prior and a specialized face restoration module. Unlike general upscaling models, it preserves individual features: eyes, nose, expression. According to benchmarks, GFPGAN achieves 95% ID similarity even with heavy blur. The source code on GitHub is community-verified.

How to Accelerate Pipeline Deployment?

We use pre-quantized models (INT8) and TensorRT for inference. For large volumes — vLLM or Ray Serve. Average quality drop from quantization is less than 1%, while speed increases 2-3x. We configure specifics for your hardware. For example, on an NVIDIA A100, p99 latency for a single 4K image is 8 seconds at batch size 1.

Comparison of face restoration methods:

What Is Included in the Work

We deliver a turnkey solution:

• Archive analysis: damage types, resolution, frame count
• Model selection and calibration for your dataset
• Pipeline development (Python, PyTorch, ONNX)
• Service deployment (Docker, Kubernetes, FastAPI/gRPC)
• Web interface for upload, before/after preview, download
• Testing on 50+ representative frames from your archive
• Documentation: API, configs, scaling instructions
• Operator training (1-2 days)
• 2 weeks of technical support post-launch

Timeline and Cost

Timelines range from 1-2 days (deployment of a ready pipeline) to 2-3 weeks (full cycle with web interface and customizations). Cost is calculated individually — it depends on archive volume, required models, and performance needs. We assess your project within one business day. Request a demo — we process 5 of your photos for free.

Our Experience

Over 12 years in computer vision and image processing. We have completed 30+ AI pipeline implementations for restoration, defect detection, and OCR. Clients include archives, museums, photo studios. Quality is guaranteed — restoration metrics are specified in the contract.

Typical Mistakes in Self-Service Restoration

• Using only one model (e.g., Real-ESRGAN alone) — faces remain blurry
• Excessive noise reduction — loss of fine details
• Skipping calibration — artifacts at tile seams
• No backup of originals

Get a consultation on the solution architecture. Contact us — we’ll discuss your archive and show a demo on your photos.

Generative AI Development: From Prompt to Production API

We often receive a task "generate a product image" — on the surface it seems simple. But behind this lies a choice between dozens of models, configuring the inference pipeline, manually solving consistency issues, integrating into the product backend, and answering why the model generates hands with six fingers in staging but not in production. Let's break down the directions we work with.

Image Generation: From Prompt to Production API

The current landscape includes FLUX.1 [dev/schnell/pro] from Black Forest Labs and Stable Diffusion 3.5. FLUX.1 [schnell] takes 4 steps instead of 20–50 for SDXL — 5–12 times faster — while maintaining higher quality. On an A100 80GB — 1.2–1.8 s per 1024×1024 image at batch_size=4.

A typical deployment issue: FLUX.1 [dev] requires 24+ GB VRAM in fp16. On A10G 24GB it fits tightly; at batch_size>1 — OOM. Solution: torch_dtype=torch.bfloat16 + enable_model_cpu_offload() from diffusers, or quantization via bitsandbytes to NF4 — minimal quality drop, memory consumption drops to 12–14 GB.

ControlNet and IP-Adapter are key tools for production tasks where controllability is needed. ControlNet with Canny/Depth/Pose maps provides structural control. IP-Adapter (especially IP-Adapter-FaceID) allows transferring character identity to generations — this is the foundation for personalized content. More about ControlNet can be found on Wikipedia.

Case study: e-commerce photography. A retailer with 8000 SKUs needed lifestyle photos for each product. Pipeline: product segmentation (Segment Anything Model 2) → background removal → inpainting with FLUX.1 [dev] using product image as IP-Adapter reference → upscale via RealESRGAN_x4plus. The generation cost is negligible compared to professional photography, providing huge savings. Throughput — 200 images/hour on 2× A100. Our extensive experience from 30+ projects ensures we select the optimal model for your task — an evaluation can be obtained upfront.

Why Is Model Selection Only Half the Battle?

Fine-tuning for a Specific Style or Character

Dreambooth and LoRA are the standard for adapting to a specific visual style or object. LoRA trains in 2–4 hours on 20–30 reference images on a single A100. Rank 16–32 is usually sufficient for style; rank 64+ is needed for precise face reproduction.

A common mistake: training LoRA too long — the model overfits to references, losing the ability to vary. Sign: at cfg_scale=7, all images look like copy-paste of references. Solved by early stopping (usually 1500–2000 steps for 20 images) and prior_preservation_loss.

For deeper customization — full fine-tuning via diffusers + accelerate with FSDP on multiple GPUs. But that already takes 40–80 hours of training and requires a truly large dataset (1000+ images).

Comparison of Image Generation Approaches

Video Generation: Which Models Are Best?

Open-weight video models still lag behind commercial ones in stability and length. Wan2.1 is the best choice for self-hosting: 14B parameters, runs on 2× A100, delivers acceptable quality for short clips.

The main pain of video generation is temporal consistency: the character changes clothing color at the third second, objects "drift." Partial solution — generation with motion_bucket_id and noise_aug_strength in Stable Video Diffusion, or using I2V (image-to-video) instead of pure text-to-video. As noted in VideoPoet research, consistency is achieved by training on long sequences.

AnimateDiff remains a working tool for short loops and motion effects on top of SD/FLUX. Not Sora, but deployable locally and predictable.

Music and Audio Generation

AudioCraft from Meta (MusicGen + AudioGen) is a production-ready stack for music generation. musicgen-large (3.3B) generates 30 s of music in ~8 s on A100. Control via text prompt and melody conditioning — you can specify a melody by humming.

Stable Audio Open from Stability AI is an alternative with length up to 47 s, better structural control (intro/verse/chorus). Deployment is similar: diffusers + FastAPI.

For voice-over and dubbing — ElevenLabs API or self-hosted XTTS v2 (see Speech AI service). For sound design and foley — AudioGen.

3D Generation: Current Practical State

3D generation has not yet reached the same maturity as 2D. But for specific tasks, tools are already working:

TripoSG and Shap-E — text/image-to-3D. Shap-E from OpenAI generates simple 3D meshes in seconds, but geometry is rough. TripoSG gives more detailed results but requires post-processing (remeshing, UV unwrapping).

Wonder3D and Zero123++ — 3D reconstruction from a single image. They work by generating multi-views (6–8 views) and then 3D reconstruction via NeuS or instant-ngp.

Gaussian Splatting (3DGS) — not generation, but reconstruction from a series of photos/videos. For product cards and real estate it's already production: 50–200 photos → 3DGS model in 15–30 min on RTX 4090 → interactive 3D viewer in browser.

What Infrastructure Is Needed for Generative AI Deployment?

Critical for generative models:

• Task queue — Celery + Redis or Ray Serve. Synchronous HTTP for image generation is unacceptable with >5 concurrent requests.
• Caching — similar prompts yield similar results. Semantic cache via embeddings (faiss + sentence-transformers) can reduce GPU load by 20–40%.
• Quality monitoring — CLIP score for text-image alignment, FID for evaluating generation distribution. Integrate into MLflow or Weights & Biases.
• Storage — generated images immediately to S3/MinIO, not on the inference server disk.

What's Included in the Deliverables

We take the project turnkey — from model selection to deployment and monitoring. The result includes:

• Model (or API integration) with performance benchmarks (latency p99, throughput).
• Pipeline documentation (prompt engineering guide, model card, dependency versions).
• Integration with your backend (REST/gRPC, queues).
• Configured monitoring (dashboards, alerts for quality drift).
• Training workshop for the team (2–4 hours).
• Warranty support for 3 months after launch — as part of our quality certificate.

We have completed 30+ projects in generative AI — this gives us the right to guarantee results.

How Is the Generative AI Development Process Structured?

1. Analysis (1–2 days): audit of current architecture, clarification of use case, selection of models and success metrics. We evaluate the project free of charge.
2. Proof of Concept (1–3 weeks): quick prototype on your data — to see real quality, not blog demos.
3. Design (1–2 weeks): pipeline architecture, infrastructure (GPU cluster/API), A/B testing plan.
4. Implementation and fine-tuning (4–12 weeks): development, LoRA/full fine-tuning, integration with queue and cache.
5. Testing (1–2 weeks): load tests, metric validation, edge-case verification (negative scenarios).
6. Deployment and monitoring (1–2 weeks): production deployment, monitoring setup, documentation.

Timeline Estimates

Integration of a ready API (DALL·E 3, Midjourney API, Stability API) — 1–2 weeks. Self-hosted pipeline with fine-tuning — 6–12 weeks. Full platform with UI, queues and monitoring — 3–6 months. The specific cost is calculated individually after analyzing your scenario.

Contact us — order a consultation, and we will select the optimal architecture for your project. Get a preliminary cost and timeline estimate for free.