Module 6 · Project 4: Marketing Image Generator (diffusion models)

OBJECTIVE: build an app that creates product and marketing images from a text description, edits existing images, and keeps a consistent brand style — and learn how image generation works and how to judge it. WHAT WE COVER (in project order): • How diffusion models create images (simple picture of the process) • Writing image prompts and the key settings: steps, guidance, seed, size • Image-to-image and inpainting (editing a part of an image) • Brand style with a ready-made LoRA (and how to train one, for research) • Searching images with text using CLIP embeddings (optional) • Evaluation: human rating and CLIP score • Responsible use: copyright, real people, labelling AI images • Deployment: Gradio app AFTER THIS MODULE YOU CAN: • Explain how text-to-image models work • Generate and edit images in code with control over the result • Judge image quality and use image generation responsibly DATA: no training data needed for the applied path — you write a set of 20 test prompts for your use case. TOOLS: Hugging Face diffusers on a free Colab/Kaggle GPU (or an image API).