At this point in 2022, several AI models have made significant progress. We can now use AI to create paintings, articles, and other content, giving rise to the AIGC industry. I tried the popular Midjourney and recorded my experience and thoughts.
Creation Principle#
Generative AI models are used to draw content described by user-input keywords. Currently, this type of algorithm is applied in the creation of 2D images, music, text content, and other fields.
The algorithm is essentially a machine learning model trained on a vast amount of image content. It more accurately identifies the content described in human text and more precisely depicts the scenes described by the text.
Practical Applications#
2D#
This year, the maturity of algorithms in multiple models has been very high. In the field of game concept art creation, AI can now be integrated into the workflow. It can address the awkwardness of relying solely on verbal descriptions during communication, which often fails to visualize ideas concretely. Production teams can subsequently use AI to generate required concept art and even create some commercially usable assets.
Currently, the image quality and detail generated by the algorithms cannot match the creation quality of top-tier game content. However, they can replace some simple creative content, such as assets used in prototypes for indie games.
Currently, we see creators already using models to generate simple forms of assets, such as in-game item icons.
From my personal perspective, this batch of item icons already meets the quality standards for commercial game assets. Previously, when creating games, one had to find similar assets created by artists for substitution. In future game development, this step of searching for resources can be saved.
The most significant impact of this is that the power of artistic creation is no longer solely in the hands of artists. Everyone in a team can use AI to intervene conceptually in the artistic creation process. The old joke “If you think you can do better, then you do it” might become a reality.
AI models have achieved a very high level of maturity in scene depiction and creation. The example image below is a concept piece I created for a Mars simulation game. In the early conceptual design phase, I was able to obtain concept designs for the city, environmental color palette, monster mountains, starry sky, vehicle convoys, and other scenes.
3D#
Game 3D art assets involve a more complex production process than 2D art assets, requiring stages like modeling -> level design -> fine-tuning. Each stage consumes significant effort from art production personnel, making 3D art assets a consistently high-cost part of game development.
The mainstream solution currently used in the game industry is PCG - Procedural Content Generation for scenes. However, implementing this solution is very costly. It requires personnel to define data ranges for scenes, and the scenes generated based on this data still need extensive detail adjustments before they can resemble real-world environments.
The efficiency of production personnel determines production efficiency. To break through this limitation, the introduction of artificial intelligence is inevitable. Microsoft’s Microsoft Flight Simulator has incorporated AI technology into its production pipeline.
Microsoft collaborated with blackshark.ai, using satellite map data combined with aerial 3D scan data to feed the AI. The AI uses this data for map annotation and recognition, then employs the annotated and recognized data for modeling, ultimately synthesizing a realistic simulation environment.
This case provides a feasible direction for the future involvement of AI in 3D art asset production.
Future Trends#
Many issues still need to be resolved for AI to be involved in production across various fields. However, it is undeniable that AI will replace some of the production techniques we currently master, forcing us to reconsider which parts of the creative process can be substituted by AI.
As the answers to these questions become clearer, solutions will emerge accordingly.
AI models will be applied to various detailed aspects after thoughtful consideration, forming specialized models. This will liberate us from previously repetitive workloads. In the new era, it is essential to master the method of conversing with AI to make it work for us.
💡 Thank you for reading! Feel free to share the article or write to me to exchange ideas.
The RSS address for this site has been updated. Please resubscribe, dear readers!