Generative media overview

Generative media means AI systems that create images, audio, video, or other media from prompts or examples—not only classify existing files.

Creative tools are mainstream. So are deepfakes and nonconsensual imagery risks (Course 29). Literacy means knowing capabilities, limits, and consent norms.

Many image systems learn to reverse a gradual “noising” process (diffusion) guided by text. Others use GANs or autoregressive token approaches. All need training data—and inherit dataset biases.

“Make a logo sketch in flat vector style” → iterate prompts → human picks and edits. The human remains the publisher of record.

Generate one image for a harmless personal project, then list: what you verified, what you still don’t trust, and whether you may use it commercially (check the tool’s terms).

⚠️ Myth: Generative tools “understand” art like a person.

✓ Reality: They model patterns in data; authorship and rights are legal/social questions.

⚠️ Myth: Watermarks alone solve misuse.

✓ Reality: Helpful signal, not a complete defense.

• Provider docs for the image tool you use (cite specifically)
• Course 29 nonconsensual-imagery overview
• Course 08 multimodality