https://editor.p5js.org/yc5965/full/ucpQWC_VM
Based on our class practice, I first experimented with adding glitch effects when no face was detected.
However, the result was very laggy because the glitch effect updated too quickly, and the visuals felt disorganized. To make it more intentional, I shifted toward creating a face collage effect.
After adjusting the size and refining some details, the results became much more satisfying.
b88ad6690b2efdbd0f8108e2e533a102.mp4
For the next steps, I was considering making the rectangles draggable to give users more freedom and control to arrange the collage in their own way.
Problem:
One recurring issue was that the MoveNet model sometimes failed to load or respond. After troubleshooting, I discovered (with ChatGPT’s help) that I needed to add a function to check whether the model was ready before starting detection. This fixed the instability and made the whole process more reliable.
function modelReady() {
console.log("✅ BodyPose 模型加载完成");
myBodyPose.detectStart(myVideo, gotResult);
}
From the participants’ side, some young people might be more open to having their works used to train models, viewing it almost like a form of taxation: you contribute, and in return, you benefit from the powerful models. However, this analogy quickly breaks down when considering those who are no longer actively creating. For example, retired artists may rely on income from their original works, yet AI models can appropriate their published material and generate countless derivatives in seconds. These artists receive no benefit from the models, while their aesthetic uniqueness is diluted. This seems deeply unfair, and currently there is little legal protection in place to address this imbalance. I would like to see future legal frameworks that safeguard the rights of such artists.
After reading the materials, I was especially inspired by Body, Movement, Language. The combination of body and speech is impressive. The live code editing is also cool and effective. Understanding the provenance of training data directly shapes how I approach my creative process. For instance, if a pose detection model was primarily trained on able-bodied dancers, it may fail to capture atypical or differently-abled movements. If I know the dataset was crowdsourced openly, I may feel more comfortable using it in participatory or community-driven projects. If instead the dataset was scraped without consent, I might avoid using it altogether—or foreground that ethical tension within the artwork. Likewise, recognizing a lack of diversity in the data could push me to frame my project as critique or intervention, rather than as a neutral application.
I am also considering the possibility of training models myself, which would shift the emphasis toward personalization and agency while allowing me to ensure that the data sources are both appropriate and filtered to align with my intentions. However, builidng, maintaining and updating datasets is resource-intensive and demanding. For this reason, I would need to choose a narrow domain to remain sustainable.