Emm Profile picture
Nov 10, 2022 24 tweets 14 min read Read on X
I've explored a reliable method to create high-quality, style-consistent #game assets w. #StableDiffusion

We're just scratching the surface here & I believe this can become a very potent creation tool (like a "@procreate on AI-steroids"🤔)

A demonstration with... spellbooks 🧵
Spellbooks are another ubiquitous asset in heroic #fantasy games, tabletop games, D&D, etc.

However (like for potions), the Unity/Unreal asset stores have limited choice, and most assets have a rather simplistic design.

Here's how #AI can solve such a situation.
At first, I generated a set of 38 "random" spellbooks, which I used to train a finetune model of Stable Diffusion (Dreambooth). It took 1 hr.

I then explored a set of possibilities with simple prompts and or limited modifiers (intricate, detailed, beautiful, 3D render...)
One word can make a difference!

In the example below, I added "realistic" (on the right). The cover is less intricate, more simplistic, and primarily leather.

Find and save your favorite modifiers (massive databases are being shared, by the way).
If you're #modding a game and need to fit into the initial art, try typing the game's name. In the example below, I added "illustration dungeons and dragons"

The style usually changes accordingly ("simpler", in that case)

You can even test a combination of two or more games :)
When I like a specific item, I don't hesitate to generate a bunch of close variants (16 variants in 30 secs).

Some variants (left) can look better than the original item (right)
Now instead of random variants, you can easily change the main color, by adding it in the prompt

Such as "purple" or "green" below
Or "blue and orange", "pink and black", etc...
Aside from colors, I often try adding details and ornaments to the item(s).

These are spell books with "a volcano", "an island", "a skull", and "an eye" on the cover. All done in seconds

Think about the time needed to draw each concept manually...
One direction I took when exploring this collection was to specifically play on the materials.

For example, "grimoire, precious metals and gemstone, stylized, natural, game icon, digital illustration, hdr"
One of these "precious metals and gemstone" variants was cool, so I generated 64 variants
... but I could do 400 variants too!

This tool is fantastic for games that need a lot of unique content (such as web3 games, where users value unique/rare assets)
If gemstones aren't the right fit, it's easy to change the materials by changing the prompt.

Let's do "grimoire, gold, silver, bronze, wood, stylized, natural, game icon, digital illustration"

The gemstones are gone, replaced by gold ornaments.
Another way to test variations is to leverage #img2img with random images or sketches on the web (or self-drawn). Below are examples of the SAME prompt (precious metals and gemstone) but different #img2img input.
Another books with skulls, also via #img2img
Now beyond shapes, colors, ornaments, or materials, you can play on "VFX". i.e. particles or potential animations that could appear around your items

Here's "magic book, enchanted, lighting sparks, fireflies". The whole batch isn't perfect, but some of the results are neat.
And I'm ending this thread with one of my favorite ideas, "a magic #vortex inside of a book":
I hope this provides a good example of properly leveraging AI by "fine-tuning" Stable Diffusion.

Bespoke AI engines will have incredible applications in #gaming. Check the other examples I recently provided (golems, potions, etc.)
And (as I had quite a few questions over the past few weeks) >> we're going to make this available to anyone that wants to build great content for games (game artists, game developers, studios, indie devs, art directors, etc)

100% online, no technical skills required

Stay tuned
PS: if you like it or want to support us, feel free to RT, we would really appreciate it!

And let us know what we should be sharing next :)

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Emm

Emm Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @emmanuel_2m

Apr 20
Designing ultra-consistent and virtually unlimited game characters in minutes with #Scenario.

Keeping it super short 😅 — because it's actually super fast🤗.
Image
Image
1/ Select your model.

You can choose from our selection of 170+ game art models on

Or, train your own to better fit your style.

I trained this model with only 12 images, originally generated using @artificialguybr's '3D Render Style' 🙏

Note: I'm intentionally training with consistent poses.

Spoiler alert 🚨: Next week, you'll be able to #upload your own models to Scenario! 🫣app.scenario.comImage
2/ Create consistent prompts.

It's best to maintain a similar structure between your prompts and captions.

I asked ChatGPT to use a format based on race, weapon, and outfit, with 50 initial prompts below👇👇.

Adjust as needed.Image
Read 10 tweets
Mar 12
Just tested Midjourney's "Character Consistency" feature (-- cref). It's promising!

I don't think it beats a good #LoRA (yet). Especially when paired w. IP Adapter.

Below are some quick tests I ran on Scenario (high-def pics below + more updates coming👇👇)

What do you think? “A blue-skinned man with large yellow eyes, bald, wearing a BJJ gi with a black belt, raising his arm in victory on a podium, pointy ears, wrinkles, highly detailed, 35 mm, award-winning photography, bokeh, studio quality”
Left: input image
Right: MJ (--cw 0)
Image
Image
Left: input image
Right: MJ (--cw 50)
Image
Image
Read 5 tweets
Feb 9
@ClaireSilver12 - (I hope you don't mind me RT this for a broader reach and to share it with more users.)

Here's an advanced use case for the IP Adapter. You can adjust or remove the steps depending on the desired output/goal. Bear with me; it's actually quite straightforward.

1 - Train a LoRA on a specific subject (e.g., character).
2 - Blend the LoRA to perfectly capture the style (e.g., comic, cartoon, oil painting, 3D...).
3 - Run inference on that "blended" model.
4 - Select an image that stands out and use it as a reference with the IP Adapter.
5 - Modify the prompt to create variations of the subject.

Let's get started 👇👇Image
1/ The first step is to train one (or more LoRA) models on a specific subject (e.g. character or object), or also a style.

The process is straightforward. I'll use the example of the "girl with pink hair" (😊🫠) that I shared before (12 training images)

Since the images were photographs, I'm creating a "mix" by adding one or two more LoRAs to the composition.

Additional LoRAs can introduce a specific style, moving away from the original photorealistic style.

You can choose platform LoRAs curated by @araminta_k and the Scenario team - See Image
Image
Read 10 tweets
Feb 1
"3D Outfit Transformation with AI"

I trained a LoRA model on , using a 3D model of an old armor, available on Sketchfab (CC0 license).

Now, I can apply it to seamlessly "dress up“ my characters. Detailed process in the following X tweets 🧵👇 Scenario.com

Image
Image
I found this nice 3D model of an ancient medieval outfit, crafted by @Blueprint_Games and shared on @Sketchfab under the CC BY 4.0 DEED license.

You can explore it here: . sketchfab.com/3d-models/old-…
Training the model is straightforward.

Simply select "New Model - Train" on . I use 9 images of the model, showcasing various angles and zoom levels, accompanied by concise captions (details below).

I selected the "Character" (= subject) preset. app.scenario.com
Image
Read 7 tweets
Jan 18
This could be the best model I've ever created for generating isometric buildings, on Scenario.

Output consistently match the style I wanted, and the model responds perfectly to (short) prompts, without any reference images needed.

It's a LoRA composition. More below.
Process: it's pretty simple.

I created a LoRA composition from 4⃣ distinct LoRA.

(i) - My own "Fantasy Buildings" LoRA
(ii) - Three LoRAs available on #Scenario: "Isometric Storybook", "Stylized Fantasy Iconic Imagery" and "Belgian School Comics".

The influence of each LoRA is below.Image
My prompt structure was dead simple... less than 10 words!

(type of building/scene), solid color background, highly detailed, centered. Image
Read 22 tweets
Jan 5
2024 will be the year of AI #influencers... Fully generated characters that can engage, entertain, and earn!

They're not just a concept anymore; some are already seeing significant revenue.

You can create one today. Learn how to build it from A to Z, including all aspects of visual storytelling, in just 30 minutes on #Scenario. 🌟👇

Just follow the steps below (or bookmark for later) 🧵Image
#GenAI will drive the emergence of virtual influencers

Take Lil Miquela, a computer-generated character with 2.6M followers, working with Burberry, Prada, or Givenchy, securing deals worth hundreds of thousands of dollars.

Or Kyra, India's pioneering virtual influencer, boasting over 200k followers and numerous brand endorsements.

More recently, Aitana López, a Spanish GenAI model created by The Clueless agency, gathered 253K followers, reportedly earning up to $11k monthly.

Other virtual stars include Lu do Magalu, Imma, Shudu Gram and this is just the beginning, here's why ⏬

Crafting an AI influencer means meticulous control for both the character details and the visual storytelling.

Maintaining the control & consistency can be daunting, and simply relying on prompts can prove to be quite challenging.

Enter : Unlock unparalleled consistency by training custom AI models, directly from your web browser, all without requiring dedicated hardware or complex local setup.Scenario.com
Read 23 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(