SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.
見出し画像

[Important Notice] 'Prompt Block Structure' has evolved to Ver. 2.0. Achieving absolute stability to ensure your character never 'breaks' again.

Hello everyone!
This is Toshi from 'AI Anime Workshop'.

※I have updated this immediately.

Thank you very much for always watching my experimental content.

Today, I would like to report on an extremely important update regarding the 'Prompt Block Structure,' which is the core technology of my project, and share one major failure behind it, while disclosing the specific prompts used.

Why was an update necessary? — The aftermath of the 'Aniko Identity Crisis'

Regarding my character, [Aniko],

The [Free Prompt Release] [Zodiac] series ended successfully, and the incident occurred while I was launching and working on the next project, 'Aniko's AI Workflow Closet'.

I will be straightforward.

The other day, I committed an unforgivable mistake by using the power of AI to generate my original character 'Aniko' as a completely different person.

This is the problematic prompt and the generated image.

Prompt
fantasy art, soft lighting, ghibli style, detailed, cinematic lighting, a young woman character 'Aniko' with soft silver bob haircut with see-through bangs, her captivating blue eyes have a galaxy-like sparkle, gentle smile, translucent skin, wearing an adventurer's outfit, close-up shot, looking slightly into the distance with a thoughtful expression, ancient ruins in a deep forest, sunlight filtering through the trees, moss-covered stones, glowing magical herbs --ar 2:3 --niji 6

Huh?? Who is this??

The limits of Ver. 1.0: Why did Aniko become a 'different person'?

It has been discovered that the Ver. 1.0 structure I have been using has a serious flaw: when a specific theme (e.g., fantasy) is specified for 'art style' or 'background,' the 'gravitational pull' of that theme's worldview is too strong, causing it to contaminate and overwrite the character's art style that should be preserved.

[Failure Example] Ver. 1.0 Prompt

Prompt
fantasy art, soft lighting, ghibli style, detailed, cinematic lighting, a young woman character 'Aniko' with soft silver bob haircut with see-through bangs, her captivating blue eyes have a galaxy-like sparkle, gentle smile, translucent skin, wearing an adventurer's outfit, close-up shot, looking slightly into the distance with a thoughtful expression, ancient ruins in a deep forest, sunlight filtering through the trees, moss-covered stones, glowing magical herbs --ar 2:3 --niji 6 --stylize 250
Japanese translation:
Fantasy art, soft lighting, Ghibli style, detailed, cinematic lighting. The young woman character 'Aniko' has a soft silver bob haircut with see-through bangs. Her captivating blue eyes have a galaxy-like sparkle, and she has a gentle smile. Translucent skin. Wearing an adventurer's outfit, close-up shot. With a thoughtful expression, looking slightly into the distance. Ancient ruins in a deep forest. Sunlight filtering through the trees. Moss-covered stones, glowing magical herbs --ar 2:3 --niji 6 --stylize 250

Ugh...

In this prompt, the powerful keyword 'fantasy art' at the beginning ignored the subsequent character definition and made the AI draw a 'fantasy-style character.' Stable character operation is impossible with this.

The solution in Ver. 2.0: Complete separation of 'Sanctuary' and 'Variable Domain'

To overcome this failure, I developed the new 'Prompt Block Structure 2.0'.

The core philosophy lies in completely separating and protecting the "region for rendering the character (= sanctuary)" and the "region for expressing the world view (= variable region)" within the prompt.

[Success Example] Ver. 2.0 Prompt

Prompt
masterpiece, best quality, official art, game character splash art, trending on pixiv, soft and luminous illustration, clean character lines, modern anime aesthetic, a beautiful young woman with short silvery-white hair and blue eyes, a gentle smile, wearing a dark green adventurer's tunic, upper body shot, looking at viewer, background is a mystical forest with ancient ruins, soft sunlight filtering through the canopy --ar 2:3 --niji 6

I wish this image would be output from the start... so here is the face-swapping procedure.

1. Generate the original image with Midjourney v7/nijiourney v6

Create the original image using this prompt.

Prompt
masterpiece, best quality, official art, game character splash art, trending on pixiv, soft and luminous illustration, clean character lines, modern anime aesthetic, a beautiful young woman with short silvery-white hair and blue eyes, a gentle smile, wearing a dark green adventurer's tunic, upper body shot, looking at viewer, background is a mystical forest with ancient ruins, soft sunlight filtering through the canopy --ar 2:3 --niji 6

Yes, it's a different person. That's fine. The trick is to choose an image with lines that are as clear as possible.

2. Ask Gemini to change the outfit

Have it change the outfit in this image.

Prompt
Please dress the woman in the first image in the outfit from the second image while maintaining the same art style and characteristics.
Japanese translation:
Please dress the woman in the first image in the outfit from the second image while maintaining the same art style and characteristics.



Since it's AI, the output results will be different every time. Let's adopt the one that looks good.

[Ver. 2.0 Improvement Points]

Establishment of an absolute art style:

At the beginning of the prompt, I defined a rendering style specifically for the character, such as official art, clean character lines, and modern anime aesthetic, and made this a [completely fixed sanctuary]. This strongly commands the AI, "No matter what happens, draw the character in this art style."

Neutralizing the influence of the background:

I changed the background specification from an immersive expression like "in a forest" to a descriptive statement like "background is a forest." This structurally prevents the world view of the background from interfering with the character's art style.

Summary: This is the full picture of the new "Prompt Block Structure 2.0"

The role of each block has evolved as follows.

1. Art Style & Quality Block (= Soup) [Completely fixed for Aniko]

Role:
The foundation that determines the overall quality of the illustration and the drawing style of the character itself. By specifying official art, clean character lines, etc., it acts as the core of the [Sanctuary] that declares to the AI, 'Aniko is not to be drawn in any other art style.'
Prompt example:
masterpiece, best quality, official art, soft and luminous illustration, clean character lines, modern anime aesthetic,

2. Character Definition Block (= Noodles) [Skeleton is completely fixed]

Role:
The skeleton that defines the unchanging physical characteristics of the character. It specifies hair color, eye color, facial features, etc. Only the clothing, 'wearing...', is the sole element that can be changed according to the theme.
Prompt example: a beautiful young woman with short silvery-white hair and blue eyes, a gentle smile, wearing a dark green adventurer's tunic,


3. Composition & Direction Block (= Chashu) [Changed per theme]

Role:
The part that instructs the character's pose and camera work. In our posts, it is basically fixed to an upper body shot, but it is a [Variable Area] that can be freely changed as needed.
Prompt example:
upper body shot, looking at viewer, thoughtful expression,

4. Background & Environment Block (= Green Onions/Menma) [Changed per theme]

Role:
The main [Variable Area] that depicts the worldview, which is the theme of the day. By thoroughly using the 'background is...' format, we prevent art style contamination of the character.
Prompt example: background is a mystical forest with ancient ruins, soft sunlight filtering through the canopy,

5. Parameters (= Pepper) [Basically fixed]

Role:
The final touch that specifies the image aspect ratio, etc. In my project, this is fixed in principle.Prompt example:
--ar 2:3 --niji 6

All future posts will be based on Ver. 2.0

With this update, all prompts published on note from now on will comply with this 'Prompt Block Structure 2.0'.

This will allow everyone to use the prompts as a reference with peace of mind and apply them to your own character management.

I hope my failures will help make your AI workflows more robust. Please look forward to the reborn 'Aniko's AI Workflow Closet'.

Having written all this, I realized a major fact.

I've spent half a day writing this, but I've just realized a major fact.

The very first prompt doesn't comply with the prompt block structure...

The cause of this is believed to be entirely due to the way I interacted with the AI.

There is absolutely no problem with using the prompt block structure as 2.0. It might be a blessing in disguise, but I am proud that I have created something good.

You can use it with peace of mind.

What is the true way to interact with AI?

Since this article has become quite long, I will end it here for now.

At a later date, I will try writing about what the true way to interact with AI is.

Thank you very much for reading until the end.

See you later~♪


いいなと思ったら応援しよう!

この記事が参加している募集