Hypercinema
Physical Computing
Computational Media
Hypercinema
Week 07-08 : Synthetic Media
Week 7 Assignment
- Find an example of synthetic media (marketing, art, entertainment, ???) Go beyond deep fakes, find novel examples if possible.
- Write a blog post (For next week Week 8) showcasing the example you found and answer the following questions:
- Why do you consider this synthetic media?
- What can you find about how this was made (if you can)? If it’s machine learning based can you find information about the data set used to train it?
- What are the ethical ramifications of this specific example (if any)?
Link to AI Emoji Generator -> https://emojis.sh/
In recent years, synthetic media - media generated or manipulated by AI and machine learning - has exploded in popularity and capability, reshaping fields like marketing, art, and entertainment.
I will focus on a unique example of synthetic media: an AI emoji generator available at emojis.sh
By exploring how this generator creates unique emojis, I will unpack why it qualifies as synthetic media, getting deeper into its technical aspects, and consider potential ethical implications.
Synthetic media is created by algorithms rather than by human hands, often based on user input or AI models. The emoji generator on emojis.sh fits this idea well. Using AI, it makes new, custom emojis by blending familiar styles and emotions. Unlike the standard emoji sets on our devices, this tool lets users create their own emojis to better match their unique needs and preferences.
Because this tool relies on algorithms instead of human designers, the emojis it creates are a direct result of AI processing. This quality makes it a clear example of synthetic media—content created by computers, not people.
The exact technical specifications of emojis.sh aren’t publicly available, but based on the functionalities offered, we can infer that it likely uses a machine learning model trained on various visual datasets. The model would need to recognize and understand key emoji characteristics - such as facial expressions, colors, shapes, and popular design conventions - and apply these to generate new emoji combinations.
The process may involve a combination of techniques such as:
•Image Generation:
Likely using image-generation algorithm, the model would be trained to synthesize emoji-like images by blending various elements like eyes, mouths, accessories, and colors.
•Style Matching:
The AI likely uses style transfer techniques to ensure the new emojis look like they belong with existing ones. By identifying the key patterns that make something look like an emoji, the AI can create images that feel familiar and recognizable.
As for the dataset used to train this model, it would ideally consist of a large set of existing emojis, likely sourced from open-source emoji repositories or images of emojis from Unicode’s official catalog. If the developers opted for additional data to improve nuance, the dataset might also include emotive icons, stylized avatars, and images depicting facial expressions to increase the emoji generator’s ability to convey different emotions accurately.
Emojis allow people to express their identities and emotions. This AI tool might miss certain cultural or personal details if the data it was trained on isn’t diverse enough, leading to limited or biased results.
2. Intellectual Property :
Most emojis are copyrighted. There’s a potential risk that the emojis created by this AI might closely resemble copyrighted designs, especially if they’re used commercially. The developers need to be cautious to avoid any legal issues with existing emoji designs.
3. Potential Misuse :
While emojis might seem harmless, there’s a risk that someone could use this tool to create offensive or inappropriate symbols. Developers may need to include filters to prevent harmful uses of the tool.
4. Privacy :
If the AI collects user input to improve results, the developers must ensure that data is kept private and used responsibly.
The AI emoji generator at emojis.sh is a fascinating example of synthetic media, showing how machine learning can expand our digital expression. By looking at this tool through the lens of synthetic media, we can see its potential to change how we communicate, while also raising important ethical questions. As AI-generated emojis become more popular, it’s essential to consider these ethical issues to make sure that these tools are both useful and responsible.
Synthetic media tools like this one bring fresh creative possibilities while reminding us to handle new technology thoughtfully.
🎬 Synthetic Media 🎬
Team : Amanda, Amelia, Jiyou, Ray
Our video explores the idea of “weirdness” and challenges how society views people and things that don’t fit in. Through the worlds of artists like Magritte, Dalí, Buffet, and Van Gogh, we question the labels society places on what it considers “strange” or “different.”
Magritte’s work invites us to rethink the judgments that limit identity. Then, in Dalí’s surreal landscapes of melting clocks and tall-legged elephants, we see a world that feels strange yet full of life, reflecting the experience of being outside of the norm. In Buffet’s harsh, rigid world, we feel the weight of society’s judgment, where “weirdness” is often met with exclusion. But in Van Gogh’s world, there’s empathy and understanding; weirdness becomes a strength, a sign of being true to oneself.
Our video encourages people to see “weird” as a special spark, a kind of courage that allows us to break out of society’s frames. Through this journey, we celebrate those who color outside the lines, leaving unique, unforgettable marks on the world.
Leonardo.Ai
I used a program called “Leonardo.Ai” to generate source images for the video. Since we needed to create images that referenced the style of specific paintings, there were many limitations in using generative AI programs. During this process,I found a generative AI tool that was very well-suited for our project, and I decided to give it a try.
The biggest advantage of this tool was its ability to provide diverse image guides. Since a specific character needed to appear continuously throughout the video, it was essential to maintain a unified style across scenes, making it feel like everything was happening within a single artwork.
This tool offered various image guides, such as character references, style references, and content references, allowing us to quickly and easily generate images in the desired style.
In this context, a content reference means referencing a specific scene you want to depict in the prompt by including an image along with the text description.
For instance, if you want an image of only the legs of a man standing on a road (up to the knees, like in the photo on the left) in the style of Bernard Buffet, you would place a real image of that composition or scene in the content reference section, and add a painting by Bernard Buffet in the style reference section.
Although it required the extra step of generating additional images for references, it was highly satisfying for achieving the desired style.
RUNWAY
ElevenLabs
We used a program called “ElevenLabs” to generate various sound effects, background music, and narration for the video. After trying several different programs, we found that this one produced the most satisfying results.
When you enter a prompt, it generates sound effects or background music that match the description.
You can adjust various settings, such as audio generation time, prompt accuracy, and the level of variation, allowing for a wide range of creative experiments.
A single prompt can generate up to four sound effects or background tracks, and even with the same settings, all four audios were varied.
It was fascinating to see how different the results could be each time, even with identical prompts and settings.
However, this large variation sometimes made it challenging, as finding a specific desired audio required going through numerous generations.
When adding narration, we wanted the speaker’s voice to reflect the theme of “weird,” so we chose a voice with noticeable ups and downs in tone, avoiding anything too monotone.
Initially, we input the script as one large block of text, but it didn’t produce the audio effect we wanted.
So, we revised the script into a storytelling format, breaking it into shorter lines with line breaks to emphasize key points, allowing the intended message to come through more clearly in the audio.
Amanda, Jiyou, Ray
Wonderful teammates who worked so hard and passionately at every step… So grateful to have had the chance to work together with you guys :)