After clarifying the uses of AI, this painter no longer feels anxious

Please note that the following content only represents a personal summary of the work process and industry thinking. It may not be exhaustive and does not involve too many technical details. If you have other ideas, please feel free to share

There are some popular science texts, students in the same major can only read parts 4 and 5.

Introduction:

The content that needs to be rewritten is: 1. Renderings of different AIGC software

2. Principle explanation and software selection

3. What exactly is AIGC industrialization

4. Give an example

5. The degree of AI intervention under different business requirements, and the requirements for equipment and personnel

6. The profit direction I see

7. The impact of artificial intelligence on me personally

01

The content that needs to be rewritten is: the renderings of AIGC software are different

The current mainstream software of AIGC includes midjourney and stable diffition.

The output graph I generated using MJ:

The output graph I generated using sd:

It can be clearly seen that the production pictures of mj are more precise and the pictures are more delicate; the quality of sd is slightly worse. The fundamental difference between the two is that the underlying framework of sd is open source, while mj is closed source.

02

Principle explanation and software selection

The essential meaning of open source is that I can freely choose any node of this technology and transform myself into a model that suits my business; so in most cases, the software suitable for industrialization will be sd instead of mj.

03

What exactly is AIGC industrialization

Under the premise that it is consistent with the business art style (painting style), AI can quickly produce a large number of drawings to solve the needs or improve the efficiency of solving the needs according to the specific needs of the project; thus reducing labor costs and improving productivity.

The available value of AIGC is proportional to the lower art requirements and the increase in quantity

There are only two core points for specific implementation:

1. A large model can be trained that is consistent with the style of the project.

2. Can be combined with software to quickly complete a variety of personalized needs.

04

for example

In my work and personal creation, in order to save time and improve efficiency, I have made many workflow attempts on AI. Due to time and privacy issues, I am not going to go into details here, but just come up with some basic ideas to give you a small example

The painting style of a small business is influenced by Korean painter rinotina. Now it needs digital characters/roles as shown in the picture below

How will AIGC be involved?

When rewriting the content, the original text needs to be rewritten into Chinese, and the original sentence does not need to appear

Train a model that matches the painting style. (This model is provided by my friend's studio Vortex Training, it is only for practice for this article, non-commercial use and no copyright issues)

1. Find about 70-100 pictures with the same style.

2. Tag.

3. Test data is run in batches until the lora with the closest and stable style is found. (loha is not recommended at present, the product parameter variable is too large)

Second step

Combine with software to quickly complete a variety of personalized needs.

The following are two workflows that I personally recommend: one is the Vincentian diagram method suitable for time-critical projects that require less thinking time; the other is the Diagram method suitable for projects with more complex thinking. . Likewise, the more you think about, the more engaged the designer will be

Rewrite the content without changing the original meaning. It needs to be rewritten into Chinese. The original sentence

does not need to appear

1. Convert needs into keywords.

2. Adjust appropriately and enter.

3. Select picture

4. I think this one is good.

You can see that there are still many details in this picture, but it can basically meet 60% of the needs. From generation to image selection, it takes about 3 hours. The remaining detailed design adjustments require manual adjustment by the designer and strong handover with the demand side.

The method flow of this set of Vincent diagrams is more about rapid issuance of a general direction. After giving the demander more choices, it is up to the designer to iteratively improve it. However, it can also be used when the demander issues reference drawings to the artist.

Rewritten as: graphically vivid:

1. If the demander is not so sure of his idea, the artist can try more sketches in the early stage to confirm the general direction.

Use the plug-in controlnet to intervene and reject unstable generation

Controlnet integrates multiple processors, such as Canny line drawing detection, OpenPose posture detection, etc., which can detect the internal information of other pictures and apply it to your own picture creation

After importing the sketch into ControlNet, the HED processor was started. (I personally think that the HED processor has more comprehensive functions, so I like to use it.)

Designers need to be highly involved and make repeated drawing revisions

This photo gives people a rather longing feeling

Such a picture can basically achieve 80% of the final effect. Its design has been basically determined through continuous revisions by the artist and AI. What is lacking is the adjustment and addition of details. This kind of workflow is more suitable for those with more ideas and more needs.

Different business sides have different needs.

Combining the needs and the degree of AI intervention, you can create your own pipeline. For example, in a game project with a certain style and a large amount of art materials, if you want to use AI to save art promotion costs, the pipeline can be:

Fixed character lora + scene lora + painting style lora or painting style large model, with any workflow of Wenshengtu or Tushengtu.

05

The degree of AI intervention under different business requirements,

Requirements for equipment and personnel

The underlying logic of AI intervention in all businesses is actually the same.

The more ideas there are and the stronger the certainty about the content, the lower the degree of AI intervention will become.

AI cannot replace designers, it always exists as a supplementary tool. For entertainment content-oriented products, the most important thing is always the content. As a perceptual product, content always requires designers to make choices and decisions after artificially sensing the demands of the market and others. The higher the requirements, the more important the content and ideas behind the art will be, and the lower the degree of AI intervention, otherwise the opposite is true.

In products with large purchase volumes and highly iterative products for the mass market, whether it is a brand or a game, the introduction of artificial intelligence is a cost-effective investment. In this type of business, artificial intelligence may not be able to reach the level of 90 points, but it can reach 70 points. As long as the model is stable and the requirements are clear, the early art efficiency can be improved by 20% or even 30%

In businesses that tend to be of high quality and require high ideas, AI for UI, UX, illustration, original painting, graphic design and other types of work is more of an improvement in local processes, and the overall improvement effect is relatively weak

AIGC currently has acceptable requirements for equipment. Local deployment of sd basically requires a 2080 graphics card or above. Cloud deployment is a good choice for small and medium-sized businesses.

The same applies to training. The current open source training packages for small models such as Lora are very mature. The main reason is that batch testing is more time-consuming and costly. For users, the learning cost of sd and mj is not very high. The requirements for aesthetics and the ability to adapt to needs are much higher than the requirements for technical capabilities. Basically, in most demand parties, 1 to 2 designers with equal proportions of technology are enough.

Multiple application directions of AI:

Пользовательский интерфейс и шрифты — это направления приложения, которые можно превратить в модели

Содержимое, которое необходимо переписать: 06

Направление прибыли, которое я вижу

Лично я считаю, что для того, чтобы быть прибыльным, в настоящее время не хватает понимания того, во что можно интегрировать ИИ, а не сосредотачиваться исключительно на самой технологии ИИ. Как я сказал в начале статьи, SD можно применять в промышленности, поскольку она имеет открытый исходный код, но технологии с открытым исходным кодом, как правило, трудно приносить прибыль.

Прямая монетизация сложна, но использовать ее в качестве пакета для приложения не должно быть так сложно. В частности, сделать это приложением уровня десерта, похожим на Miaoya Camera (низкий спрос со стороны пользователей, большое количество пользователей, одна функция) или бизнес-приложением (управление пакетами) для предприятий

Речь идет скорее о достижении прибыльности путем преобразования спроса в применимые возможности.

07

конец

На самом деле, будучи художником, автор очень волновался, когда впервые столкнулся с ИИ, он даже думал, что останется безработным, и хотел сменить направление своей карьеры, чтобы избежать ИИ. Но благодаря неоднократным попыткам обучения и исследованиям я обнаружил, что появление ИИ значительно снижает сложность создания красивых фотографий, чем раньше. При этом требования рынка к повествованию, эмоциональности и стилю снимков также значительно возрастут.

В последующей долгой карьере требования каждого к контенту возрастут. Существенное изменение заключается в том, что мне нужно научиться предсказывать, какой вид искусства понравится рынку в следующий раз, и масштаб инноваций, которые примет аудитория. Может ли контент, который я создаю, быть немедленно воспринят аудиторией и может ли он найти отклик у других... это стало моим новым профессиональным предложением

Каждая картина, которую я рисую, — это не картина, а часть содержания. Одна красивая картинка бесполезна, но если это сотня красивых картинок, соединенных вместе, сопровождаемых историями, сеттингом, геймплеем, моделями и анимацией, то это становится началом нового контента.

Что для меня является сутью контента?

В нескольких вселенных Человек-Паук может быть чёрным или белым, человеком или животным, а стиль рисовки тоже будет разным. Но пока его укусил паук и он забрался на вершину высокого здания, он — Человек-Паук. Точно так же, как сюжет в «Человеке-пауке: Через вселенные»

Мне нравится Человек-Паук

Содержимое, которое необходимо отредактировать: Редактор контента Game Grape Recruitment

Нажмите «Прочитать исходный текст», чтобы узнать больше

Рекомендуется к прочтению

Лилит Десять лет | Как оценить Tencent | MiHoYo Двенадцать лет

Dragon Fist Storm Defeat | Назад в будущее: 1999 | Новые продукты в ближайшие три года

Диалог с Ло Цзысюном | Диалог с Ма Сяои | Диалог с Чэн Лянци

Установить звездочкой (Процесс описан выше)

Вы можете получать push-уведомления и оформлять обложки как можно скорее!

The above is the detailed content of After clarifying the uses of AI, this painter no longer feels anxious. For more information, please follow other related articles on the PHP Chinese website!