Relight

RELIGHT brings studio-grade, cinema-quality lighting to any image — powered by AI, guided by your creativity, delivered instantly.

Comment

🚀 RELIGHT — Reimagining Light Itself

An AI-Driven Cinematic Lighting Intelligence Assistant

💡 Inspiration

Lighting is the soul of visual storytelling.

But mastering it requires:

  • Studio equipment

  • Technical expertise

  • Complex editing workflows

  • Hours of manual masking

So I asked:

“What if lighting could be redesigned automatically — intelligently — like an artist?”

That question became RELIGHT — an AI-powered lighting assistant built on IC-Light, **Real-ESRGAN, and **GPT-4o-mini.

RELIGHT doesn’t “edit” light.
It understands it.

✨ New Added Point
Today, companies spend thousands of dollars on professional lighting setups just to showcase their products. RELIGHT allows brands to generate studio-grade product lighting instantly — saving time, money, and production effort while still achieving advertising-quality visuals.

🎬 Elevator Pitch

“RELIGHT brings studio-grade, cinema-quality lighting to any image — powered by AI, guided by your creativity, delivered instantly.”

🎯 Hero Features

Each feature replaces a tedious editing workflow with a single seamless action.

⭐ FEATURE 1 — AI-Recommended Lighting (GPT-4o-mini)

Sequence:

Upload main image → Click “AI Recommendations” → GPT suggests 3 lighting styles → Click “Apply” to relight.

Pain Points

  • Users don’t know which lighting will look best

  • Hard to understand lighting terminology

Solution

RELIGHT sends the image (as base64) to GPT-4o-mini.

GPT analyzes:

  • Shadows

  • Highlights

  • Exposure

  • Mood

  • Color ambiance

Returns three expert lighting suggestions (max 3 words each).

Example Outputs

  • “Soft Rimlight”

  • “Moody Sideglow”

  • “Warm Spotlight”

Outcome

  • Fast professional lighting guidance

  • Great for beginners & professionals

⭐ FEATURE 2 — Background-Aware Relighting (IC-Light FBC)

Sequence:

Upload main image + upload background → Click “Relight” → IC-Light matches the subject to the scene.

Pain Point

Cut-out subjects on new backgrounds often look fake.

Solution

IC-Light analyzes:

  • Ambient light direction

  • Color temperature

  • Shadow falloff

  • Contrast levels

Then automatically relights the subject to match the new background.

Outcome

  • Natural, realistic composites

  • Perfect light matching for professional edits

⭐ FEATURE 3 — Manual Text-Based Relighting (IC-Light FC)

Sequence:

Upload image → Type your lighting style → Click Relight.

Pain Points

  • Traditional lighting needs masks, layers, and pro skills

  • Hard to manually recreate lighting moods

Solution

Type anything like:

  • “golden rimlight”

  • “soft cinematic glow”

  • “cold sidelight”

IC-Light applies photorealistic text-based relighting.

Outcome

  • Beautiful lighting results

  • Studio-quality in seconds

⭐ Interactive Light Gizmo

Move the light. Watch the magic.

Pain Point

Lighting angles are hard to describe in text.

Solution

A draggable light pointer converts its position into IC-Light parameters.

Outcome

  • Intuitive visual control

  • Full creative flexibility

⭐ High-Resolution Enhancement (Real-ESRGAN)

Clarity that feels real.

Pain Point

AI images often look soft or low-res.

Solution — Real-ESRGAN x4plus

  • High-detail super-resolution

  • Minimum 1080p

  • Sharp edges

  • Optional face enhancement

Outcome

  • Professional-grade output

  • Perfect for production, printing & design

🔧 How We Built It

🏗 High-Level Architecture

UPLOAD
→ RMBG (Background Removal)
→ RELIGHT (IC-Light FC or FBC)
→ GPT Lighting Advisor
→ UPSCALE (Real-ESRGAN)
→ FINAL OUTPUT

💻 Backend Stack

  • Python 3.10

  • PyTorch 2.x (CUDA)

  • Diffusers 0.27.2

  • Transformers 4.36.2

  • BRIA RMBG 1.4

  • SafeTensors

  • Flask API

Models Used

  • IC-Light-FC → text-conditioned

  • IC-Light-FBC → background-conditioned

  • Real-ESRGAN-x4plus → super-resolution

  • GPT-4o-mini → lighting intelligence

🎨 Frontend Stack

  • HTML + CSS + JS

  • Drag-and-drop interface

  • Draggable light gizmo

  • Live preview

🧠 System Flow

1. Image Upload

  • JPG/PNG/WebP

  • Auto-resize

  • Instant preview

2. Background Removal (RMBG)

Clean alpha-matte extraction.

3. Relighting Engine (IC-Light)

  • Subject encoding

  • Lighting embeddings

  • Diffusion refinement

4. GPT Lighting Advisor

Intelligent lighting suggestions.

5. Upscaling (Real-ESRGAN)

  • 1080p minimum

  • Sharp, clean details

⚠ Challenges

  • VRAM limits on 8GB GPUs

  • Fast FC ↔ FBC switching

  • Keeping inference < 12 sec

  • Designing intuitive gizmos

  • Maintaining realism

🏆 Achievements

  • Complete working pipeline

  • Accurate text & background relighting

  • GPT-powered lighting suggestions

  • Smooth 1080p upscaling

  • Modern, clean UI

📚 Key Learnings

  • Lighting in diffusion models is sensitive

  • Multi-model pipelines need optimization

  • GPT greatly boosts creativity

  • Background matching is essential

🔭 What’s Next

Immediate

  • More lighting presets

  • UI animations

  • Real-time preview

Advanced

  • 3D relighting (depth + normals)

  • Mobile version

  • Public API

🛠 Built With

  • PyTorch

  • IC-Light

  • Real-ESRGAN

  • RMBG

  • Diffusers

  • Transformers

  • Flask

  • JavaScript

  • GPT-4o-mini

🌟 Why It Matters

RELIGHT democratizes lighting for:

  • Photographers

  • Designers

  • Filmmakers

  • Creators

  • Everyday users

  • Brands & advertisers wanting studio-quality product shots without expensive lighting setups

It turns complex lighting workflows into pure creative expression.

Built With

  • diffusers
  • flask
  • gpt-4o-mini
  • ic-light-fbc
  • ic-light-fc
  • javascript
  • real-esrgan-x4plus
  • rmbg
  • transformers
Share this project:

Updates