🎨 5 Epic Ways to Use Relight

Ai5 - The AI Art Newsletter - Logo

Welcome, Explorer 🤖

Do you remember the first time you used ChatGPT?

I remember getting the same feeling as I would when playing a new video game as a kid.

Entertainment, excitement, imagination.

I think that’s AI in a nutshell.


Today’s Ai5:

  • 🖼️ Magnific AI’s New Relight Feature
  • 🔦 Crazy Real-Time Diffusion Technique
  • 🤓 5 Go-to Midjourney Permutation Prompts (download)
  • 🎦 Amsterdam’s AI Film Festival
  • 🦾 The Rise of Generative 3D

Prompt of the Day 🎨

A whimsical sketch of a beautiful lady on a chalkboard, her long hair behind forms a raging dragon that chases her. Shadows and light play across the surface, chalk, blackboard, sketch, texture, contrast --chaos 10 --ar 16:9 --style raw --stylize 90
A whimsical sketch of a beautiful lady on a chalkboard, her long hair behind forms a raging dragon that chases her. Shadows and light play across the surface, chalk, blackboard, sketch, texture, contrast –chaos 10 –ar 16:9 –style raw –stylize 90

Got an epic prompt? Email me to get featured. 🌟


Magnific AI’s Relight Drops 🖼️

​Magnific​ is an AI-powered image upscaling and transformation platform.

This week they’ve dropped a brand new feature called Relight. It’s got some pretty epic use-cases!

Here’s 5 ways to use Relight 👇

Color Grading and Shadowing

Relight Example 01 - Color grading, lighting, and scene change example
Color grading, lighting, and scene changes (​see original​)

Hand Sketch to Image

Relight Example 02 - Sketch to Photograph
Converting a hand sketch to a photograph quality image (​see original​)

Reference Image Lighting

Relight Example 03 - Reference Image Lighting
Combine your image with a reference image to capture the lighting.

Lightmapping

Relight Example 04 - Lightmapping
Image 1 + Image 2 = Image 3 (this feature is my favorite)

Product Photography

Relight Example 05 - Product Photography
Make super fast iterations (​see original​)

You can find heaps more examples like this in ​this thread​ from Magnific’s founder.

You can ​try Magnific​ here.


Crazy Real-Time Diffusion 🔦

Real-time diffusion describes AI image generation fast enough that it’s considered to be real-time.

The gif below shows an example of real-time text-2-image diffusion using Leonardo’s Realtime canvas. 👇

Text2image real-time diffusion example.
Realtime text-2-img diffusion using ​Leonardo’s​ Realtime Canvas.

It’s using the text prompt to preview images in real-time.

​Hugues Bruyere​ has taken the concept to another level using an image-2-image approach.

By blending video camera footage with AI generated images using real-time diffusion, he’s created some truly epic shots.

Below you can see Hugues blending torch light with an origami prompt in Stable Diffusion. 👇

Real-time diffusion example
Realtime Diffusion using Torch Light (​see original​)

This is one of many examples Hugues shares on X, I highly recommend you give him a ​follow​!

This one shows what milk, acrylic ink, and real-time diffusion can do. 🤯

Real-time diffusion example
Real-time Diffusion using Ink (​see original​)

Midjourney Permutation Prompts 🤓

Permutation prompts are one of my favorite tricks in Midjourney. I tell everyone about them.

Ai5 meme - Do you use permutations? By Ai5 Newsletter
Ai5 Admin.

I figured I’d share my favorite 5 permutations I use to get faster results in Midjourney.

1️⃣ Old Faithful: Good for generating quick albums over a full range of stylize and chaos values.

–s {100, 200, 600, 1000} –c {1, 50, 100}

2️⃣ Balancer: Good if your prompt has two distinct themes. Can help balance keywords in the prompt faster.

[phrase/word]::1 [phrase/word]::{1, 2, 3}

3️⃣ Styler: Good if your generations are close but not perfect. Generate a library by slightly varying stylize.

{–s {80, 90, 100, 110, 120}, –style raw}

4️⃣ Cycle Raw: Basically the same as above but this one cycles raw as it works through the stylize values (took me ages to work out how to do this)

–s {80, 90, 100, 110, 120} {–w 0, –style raw}

5️⃣ Shotgun: Good if you’re feeling lazy or want unexpected results. Generates 144 (36×4) images in total.

–s {100, 200, 600, 1000} –c {1, 25, 50} –w {1, 1500, 3000}

Note: Balancer isn’t working in Alpha right now, I submitted a ticket! Discord works.

Ai5 Newsletter - Midjourney Permutation Prompt Cheat Sheet
​Download​ the high-def version.

Amsterdam’s AI Film Festival 🎦

AI film festivals are popping up everywhere these days with another one going down last week.

The ​AI Film Fest Amsterdam​ was the first annual event in Amsterdam for AI filmmakers.

Here’s some of the 2024 finalists. 🥇

👆 Dragged Holidays: Best short film, runner-up best cinematography.

👆 Black Hole: Best sound design, Best cinematography.

👆 e^(iπ) + 1 = 0: Best hybrid film (this one’s my favorite so far)

The stories are so good! Once the visuals catch up (like when Gen-3 is available) I think AI films are gunna blow up.

A full list of entrants can be found on the ​website​.


The Rise of Generative 3D 🦾

Generative AI continues to break into new ground. Let’s take a closer look at what generative 3D is, and what it can do. 🔍

Image to 3D example
Turning a Midjourney image into a dancing 3D model (​see tutorial​)

Generative 3D uses AI to create 3D digital models and environments from text prompts, images, and videos.

Applications for gen-3D include:

  • ​Generative design​ is being explored by all the CAD software companies. Instead of Engineers manually modeling parts, they’ll enter parameters and have AI generate concepts for review.
  • ​3D animation​ is becoming accessible to anyone via gen-3D. This eliminates the need for motion capture or manually animating with keyframes.
  • Generative models for ​high-quality 3D assets​ are an active area of research. These consist of text-to-3D, image-to-3D, and models that generate PBR textures for computer graphics.
PBR Texture generation example
PBR texture generation (​see original​)

These examples are only just scratching the surface of the research happening behind the scenes in gen-3D. 👀


Snack Sized 5 🍪

1️⃣ The big 3 music labels ​Sue Udio and Suno​ for copyright.

2️⃣ Someone created a ​Dave Chappelle deepfake​ using comedy written with Claude Sonnet.

3️⃣ This Swiss company is ​growing brains​ in the lab for computational work, citing massive energy savings.

4️⃣ ​Cool application​ of Midjourney and frame interpolation to create animations.

5️⃣ Toys ‘R’ Us debuts an ​advertisement​ created using OpenAI’s Sora.


Help us grow by sharing Ai5 with a friend. I’ll give you a cookie. 🍪

SUBSCRIBE to Ai5

Join the conversation on X.

👈 Read Previous Issue | Read Next Issue 👉

Keep reading on Rareconnections.

If you have any special requests, tips, or you just want to talk about AI, email me: matt@rareconnections.io