AI for the Advertising Industry! Stable Image Ultra Techniques (1): Adding Fashionable Personality and Impact with "Neon-Punk"

I'm Hanegoro, and I'm happy to be working with AICU.
I work as a producer, applying my many years of experience in advertising video and virtual live events to the era of "creating the creators."

Recently, the technical team at AICU AIDX Lab generated 1,800 images using the official Stable Diffusion API. I would like to explore how to use these as visuals for the advertising industry over several installments, including techniques for selecting which images are actually usable.

Stable Diffusion - Generate API https://platform.stability.ai/

Nominal characteristics of each model

Stable Image Ultra 1-megapixel generation (hereinafter, Ultra)
Stable Image Core 1.5-megapixel generation (hereinafter, Core)
Stable Diffusion 3 Large 8-billion parameter model (hereinafter, SD3L)
Stable Diffusion 3 Medium 2-billion parameter model (hereinafter, SD3M)

Top left is SD3L, top right is Ultra, bottom left is a composite of SD3L and Ultra, and bottom right is 3M. For Ultra and SD3L, styles are specified via text prompts rather than the API, and they tend to produce similar images with a common seed.
The technical team's impression was, "We would like you to review the commercial value of comparing photorealistic styles like 'cinema' and 'analog film' between SD3L and Ultra," so I will take on that challenge.

Evaluation Method

Just like evaluating movies or VFX, we will play back the generated stills in a sequence (a "rush") using the following tool.

❏ Video playback tool: Keyframe MP https://zurbrigg.com/keyframe-mp

We will check the rushes one by one and classify them based on the following criteria.

❏ The evaluation criteria are in the following 3 stages:
[Bad] Inconsistent art style within the frame, unpleasant anatomical failures, middle fingers raised (commercial viability, gore, NSFW).
[Good] Commercially viable, catchy composition, no obvious anatomical failures, worth considering.
[Best] Ready for client presentation, usable as a top image, requires almost no correction as an illustration.

The evaluation targets are the following realistic styles among the diverse range of styles.

The "style" here is very simple: just specify Ultra in the API and write the style in the prompt.

For detailed generation methods, please refer to the past AICU media article "#LearnAIin10Minutes".
(I will introduce a script that can generate these in bulk at the end of the series.)

Neon-Punk Style

The Neon-Punk style is rich in both Best and Bad results. It has impact, a fashionable feel, and personality, so I would like you to try it first.

Best

Bad

Reasons for Bad... Despite having high-quality images that could be the Best, they fall short due to issues with commercial viability, skin exposure, or unpleasant anatomical flaws in the fingers, especially in cases where a middle finger is being raised.

Close

Very close

The composition, direction, and expressions are good, but the fingers are flawed.

(Since there are cases where a middle finger is raised, these are examples that are still viewable among them.)

That concludes the introduction to "Neon-Punk," a style that is the most unique and rich in both Best and Bad examples, offering impact, a fashionable feel, and individuality.

This is a style I would like you to try out first with various prompts.

You can check all other images in this album.
Google Photo https://j.aicu.ai/SD3GP

Since we're at it, I'd like to hold an "SD3 Beauty Contest."
I will not claim copyright on the images in the album above, so if you find a good image, please share it with the hashtag #SD3GP .

I'll have the AICU AIDX Lab make a cool male version too!

If you, the readers, have any requests, I would appreciate it if you could mention or DM us at X@AICUai.
For technical inquiries regarding image generation using Stable Diffusion, VFX, or advertising technology, please contact AICU AIDX Lab at sai@aicu.ai

<To be continued>
❏ AI for the Advertising Industry! Stable Image Ultra Techniques (2) Exploring Tone and Manner with the Surprisingly Unique "Origami"