[CSCE 470 / CSCE 870]  Generative AI and Design  (Spring 2025)

"Generative art refers to any art practice where the artist creates a process, such as a set of natural language rules, a computer program, a machine, or other procedural invention, which is then set into motion with some degree of autonomy contributing to or resulting in a completed work of art"

CSCE 470 / 870: Generative AI and Design [Computer Graphics]

Credits: 3 Credit Hours

Schedule: Tuesdays & Thursdays | 11:00 AM – 12:15 PM

Location: Nebraska Hall, Room W193

Instructional Team

• Instructor: Dr. You-Jin Kim — ykim65@unl.edu

  • Assistant Professor, School of Computing  |  Director, Dynamic Reality Lab

• Project Assistants:

  • Lijun Mao (PhD Student) — lmao2@unl.edu

Office Hours

• Walk-in Hours: Tuesdays & Thursdays, 12:15 PM – 1:15 PM (immediately following class)

  • Location: Nebraska Hall, Lab W192

• By Appointment: Available upon request.
  Note: Please submit appointment requests via email at least 48 hours in advance.

Course Format

While this course introduces the foundational research papers that have historically defined and shaped the field, its primary objective is to bring reality to the fundamentals of generative artificial intelligence and neural network training, operating under the core premise that true mastery of deep learning cannot function without a comprehensive, practical understanding of local infrastructure execution. To achieve this, the class dives into a rapid background of recent advancements and mathematical frameworks before transitioning into a complete, hands-on project focused on implementing and deploying the latest open-source models within a working repository designed to gauge the current capabilities of state-of-the-art diffusion architectures and open-source large language models. Students will work directly in the computing lab with an extensive suite of hardware available to try out and develop for, including advanced, high-end workstations equipped with two NVIDIA RTX A6000 GPUs with 48GB of VRAM, alongside dual NVIDIA Quadro RTX 5000 cards with 32GB of VRAM, and six NVIDIA RTX 5070 Ti cards featuring 16GB of VRAM. Working together in the lab, teams will build functional pipelines specifically to challenge new creative design workflows, testing their optimization concepts to see if a deeply specialized, locally tuned open-source implementation can outsmart and outperform massive, jack-of-all-trades proprietary models like OpenAI's GPT-4, Midjourney, or DALL-E 3 on targeted spatial tasks. Consequently, enrollment requires students to be entirely comfortable collaborating closely, using GitHub and robust version control to maintain a working repository, thoroughly documenting their development process, and formally presenting their completed engineering outcomes.

Class Participation and Attendance 

Students are responsible for actively and thoughtfully contributing to these discussions and critiques. Students are also responsible for providing feedback on reading reflections and course assignments by other students.

Attendance and participation are graded on an additive basis, similar to standard assignment submissions, rather than a system of deductions. For participation, the quality of your contribution is far more important than quantity; it is entirely possible to speak frequently but receive a low grade if the contributions lack substance. Active participation within your specific team discussions is also heavily weighted. Class participation is graded on a diminishing marginal returns basis. Regular attendance is expected, as missing class directly harms the quality of the discussion and detracts from the experience of your peers. Consequently, missing 3 classes will result in a considerably lower overall grade for the course. If an absence is unavoidable, you must clear it with me prior to the start of class. 

Diffusion Models for Generative Design

Diffusion models have redefined generative AI, establishing a premier paradigm for computer vision and image synthesis that improves upon the stability and performance of Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and autoregressive models. This course explores the mechanics of diffusion-based neural networks, tracing their evolution from the mathematical foundations of forward Gaussian noise injection to the reverse denoising process used to synthesize high-fidelity assets from textual and structural prompts. Students will analyze state-of-the-art frameworks, including Stable Diffusion, DALL-E, Midjourney, and Imagen, to master the core principles of text-to-image workflows.

Beyond theory, this curriculum emphasizes the end-to-end engineering pipelines necessary to navigate today’s generative ecosystem. Participants will learn to parse and evaluate the vast landscape of open-source models, including Large Language Models (LLMs) like LLaMA and DeepSeek, filtering by parameter scale, task alignment, and licensing. Students will gain practical experience installing, configuring, and optimizing Stable Diffusion on local architectures, customizing latent spaces, specialized modules, and plugins to tailor outputs to precise artistic and functional constraints. Furthermore, the course trains students to reverse-engineer pre-generated components and 3D digital assets, breaking them down into their underlying pipeline stages and architectural origins.

A core pillar of the course is transforming AI from a passive automation engine into an interactive co-pilot for rapid ideation, structural design, and workflow optimization. To prepare participants for professional and academic leadership, the curriculum enforces rigorous engineering standards. Students will build and maintain a professional GitHub repository to host their specialized pipelines, benchmark programmatic workflows, and practice strict version control. Concurrently, final exploratory research, benchmarks, and novel architectural pipelines will be formalized and authored to meet the formatting and structural submission guidelines of IEEE, ACM, and arXiv. By bridging raw creative exploration with technical execution, this course equips students to contribute meaningfully to both industrial pipelines and academic literature.

Course Goals

• Design and evaluate generative computer vision and deep learning models optimized for local workstation hardware to ensure full control.

• Model and implement bio-inspired computational frameworks like genetic algorithms to simulate organic growth and complex physical topologies.

• Execute real-time 3D neural rendering and latent diffusion pipelines to construct and dynamically modify expansive digital design worlds.

• Design, test, and analyze automated software pipelines with open-source LLMs to customize generative systems for specific aesthetic needs.

• Construct production workflows that systematically account for dataset bias and copyright frameworks to verify creative technology practices.

Selected Related Works

• 1989 | Genetic Algorithms in Search, Optimization, and Machine Learning | Addison-Wesley

• 2016 | You Only Look Once: Unified, Real-Time Object Detection | IEEE/CVF CVPR

• 2022 | High-Resolution Image Synthesis with Latent Diffusion Models | IEEE/CVF CVPR

• 2023 | 3D Gaussian Splatting for Real-Time Radiance Field Rendering | ACM SIGGRAPH

• 2023 | DreamFusion: Text-to-3D using 2D Diffusion | IEEE/CVF ICCV

• 2023 | LLaMA: Open and Efficient Foundation Language Models | Meta

• 2023 | QLoRA: Efficient Finetuning of Quantized LLMs | NeurIPS

• 2023 | Do Models Copy? Evaluating Data Mimicry in Latent Diffusion Models | IEEE/CVF CVPR

• 2024 | DeepSeek-V3 Technical Report | DeepSeek-AI

• 2024 | Training-Time Attribution and Use-Time Valuation in Generative AI Creative Work | IEEE/CVF CVPR

• 2024 | Text-to-3D Generation with Rich Texturing and High-Fidelity Geometry | ACM SIGGRAPH

• 2024 | Scaling Rectified Flow Transformers for High-Resolution Image Synthesis | IEEE/CVF CVPR

• 2024 | VR-GS: A Physical Dynamics-Aware Interactive Gaussian Splatting System in Virtual Reality | ACM SIGGRAPH

• 2025 | DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | Nature

• 2026 | Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models | ICLR

Academic Integrity 

Honest Academic Research: Failure is part of every great research, embrace the error, document what you have learned from troubleshooting. When moving forward at a fast pace, we often overlook to cite the other’s work your project is built on top of. Integrity in academic research is essential for a valuable educational experience that being said, presenting your system and method as it is - is the first step to stay grounded to the truth and honesty. In this class, just like the real academic laboratories, you are permitted to get help from peers, getting solutions from an outside source but make sure to cite and reference how you got to your solution.  Each of you is capable of composing brilliant and original work that can meet the highest academic standards do not resort to plagiarism! To learn more about proper citation use, recognizing actions considered to be cheating or other forms of academic theft, and students’ responsibilities, please check Here. 

That being said, academic dishonesty is an assault upon the basic integrity and meaning of a University. Cheating, plagiarism, and collusion in dishonest activities are serious acts which erode the University’s educational and research roles and cheapen the learning experience as well as the value of one’s degree. All code from other sources should be properly credited regardless of your intention. When using other people’s code: include the source, author, and which functions from that code are you using or your own code builds upon. Some assignments prohibit the use of external code, some others will allow it moderately, this will be communicated when assigning coding exercises. Using "borrowed" code without citing source will be considered plagiarism. When in doubt, ask.

Course Communication & Syllabus Policies

Primary Communication Hub

All primary course communication will be conducted via Discord. Students are required to check our course Discord server daily (at least once every 24 hours) for updates, discussions, and reminders.

Dynamic Syllabus & Student Responsibility

The instructional team reserves discretionary authority to modify this online syllabus at any time during the semester to adapt to the pace of the class. Because this document is dynamic and subject to updates, students are strongly advised not to print or save offline copies, as they can quickly become obsolete and outdated.

Except for emergency adjustments, any changes to policies, deadlines, or assignments will be announced at least 10 days prior to taking effect. You are responsible for regularly reviewing the live online syllabus directly to ensure you are following the most up-to-date schedule, policies, and requirements.

Emergency & Class Cancellation Procedures

In the event of an unforeseen circumstance or class cancellation:

1. A notice banner will immediately be placed on the main page of the course website.

2. A simultaneous alert will be broadcast in the Discord #announcements channel.

In the event of campus-wide disruptions or the activation of university instructional continuity plans, students are responsible for actively monitoring both the course website and Discord daily to stay aligned with course expectations.

Academic Honesty & AI Usage Policy

Academic integrity is a fundamental value of the University community, and students are expected to complete their work with honesty.

• AI Usage & Attribution: The unauthorized use of artificial intelligence to complete coursework is a violation of the Student Code of Conduct. If you utilize AI tools (such as GitHub Copilot, ChatGPT, Claude, etc.) to assist you step-by-step in your development process, you must explicitly document it.

• Required AI Log: For any assignment or code submission utilizing AI, you must include a line-by-line or section-by-section AI usage log indicating:

  • Which tool/IDE extension was used (e.g., VS Code + GitHub Copilot v1.2 or ChatGPT 4o).

  • The exact prompts or assistance requested.

  • Where exactly in the codebase or assignment the AI output was integrated.

Academic integrity is a fundamental value of the university community, and students are expected to complete their work with honesty. The unauthorized use of artificial intelligence to complete coursework is a violation of the Student Code of Conduct. If you utilize generative AI tools or IDE extensions (such as GitHub Copilot, ChatGPT, Claude, etc.) to assist you step-by-step in your development process, you must explicitly document it.

For any assignment or code submission utilizing AI, you must maintain a strict documentation standard:

In your codebase (applicable to almost every assignment), any section that includes AI-generated logic must be explicitly annotated with code comments. You must use inline comments (//) to state the exact prompt used, which tool was utilized, and which specific lines or blocks of code were copied. If this inline attribution is missing, it will be considered unauthorized AI usage and treated as stolen work.

Furthermore, if your code or project is heavily or entirely built using AI assistance (such as Claude), relying solely on inline code comments is insufficient. In these scenarios, you must provide a comprehensive, step-by-step documentation file in Word or PDF format. This document must include a chronological sequence of screenshots mapping out your entire development process. Each screenshot must clearly display the prompt you entered, the tool's specific output, and a brief note explaining how that step was integrated into your system. Failure to provide this step-by-step visual log for AI-dominant projects will be treated as a violation of the academic honesty policy.

• Sanctions: All students enrolled in any computer science course are bound by the School of Computing academic integrity policy: cse.unl.edu/ugrad/resources/academic_integrity.php. Do not plagiarize writing or code, and properly cite all sources. Any cheating, unauthorized AI use, or plagiarism will be reported to the Department Chair and your Dean, and will result in academic sanctions, including an automatic grade of “F” for the assignment or the entire course, alongside referral to the Office of Student Conduct & Community Standards.

Industry Standards & Professional Communication

This is an advanced, upperclassman course designed to replicate a elite professional environment. We envision and train for quality standards that fit the expectations of top-tier industry leaders and research institutions, including Google XR, Magic Leap, Niantic, Bell Labs, Microsoft Research, Apple Vision Research, Amazon Research Robotics 126, NASA, Tesla, Waymo, and SpaceX.

To prepare for this level of professionalism, the following communication rules apply:

• Student Communication Responsibility: It is the student’s strict responsibility to check the course Email, Discord, and Schedule Page at least once per day.

• Professional Email Etiquette: All emails must follow a correct, professional format (appropriate subject line, formal greeting, clear concise body text, and professional sign-off). Please keep in mind that the instructor/TA response time is up to 48 hours. Plan your inquiries accordingly.

• A Note on Kindness & Mutual Respect: We maintain a culture of mutual respect. If you exhibit a disrespectful attitude towards your peers or the instructor—including being intentionally disruptive, sleeping in class, or being clearly inattentive—I will kindly pull you aside or address it on the spot once. Consistent professionalism is required to remain in good standing.

Workload

Per the UNL 1:2 Credit Hour Rule, a 3-credit course requires a baseline minimum of 6 hours of independent work per week.

⚠️ Portfolio Expectation: Because this is an advanced, project-based course—not a fundamentals class—you should realistically budget 9 hours per week of independent development. This is your opportunity to move beyond basic concepts and build an self-driven rapid prototype that you will be proud to showcase to future employers. 

Unlike lecture-heavy classes, your outside hours will be highly active and technical:

• Local Environment Configuration & Pipeline Debugging: Open-source AI development is notorious for dependency conflicts and hardware constraints. Troubleshooting custom Python pipelines, optimizing CUDA allocation, or resolving local model installation errors takes focused, uninterrupted blocks of time.

• Managing Compute & GPU Wait Times: Local AI rendering, model fine-tuning, and 3D asset generation take substantial time to execute. You must budget your hours carefully to account for these heavy processing and compilation windows—you cannot complete these projects last-minute, as your GPU simply will not run fast enough.

• Building & Automating Design Projects: We target standards expected by high-end design studios and bleeding-edge tech research. High-fidelity 3D assets, custom-trained architectures, and integrated spatial environments require consistent weekly iteration.

• Hardware Optimization & Local Inference Testing: Testing requires hands-on workstation setup—monitoring VRAM limits, optimizing local inference speeds for Llama and DeepSeek, and evaluating high-resolution neural rendering outputs in real-time.

💡 The Bottom Line: Treat these outside hours like a locked-in, weekly lab shift. Schedule them into your routine just like an in-person class meeting.

Services for Students with Disabilities (SSD)

UNL is committed to providing flexible and individualized accommodations to students with documented disabilities.

• Registration: To receive services, students must register with the Office of Services for Students with Disabilities (SSD).

• Location/Contact: 117 Louise Pound Hall; Phone: 402-472-3787; Email: ssd@unl.edu.

• Student Responsibility: Students are encouraged to contact the instructor for a confidential discussion of their individual needs for academic accommodation after providing documentation from SSD.

Mental Health & Well-being

UNL offers several resources to support students struggling with stress, depression, anxiety, or adversity.

• Counseling and Psychological Services (CAPS): A multidisciplinary team that helps students explore their feelings and learn helpful ways to improve emotional well-being. Call CAPS at 402-472-7450 at any time (24/7), including after hours.

• Big Red Resilience & Well-Being (BRRWB): Provides one-on-one well-being coaching to enhance resilience, self-compassion, and positive experiences. Reach them at 402-472-8770.

• Emergency: In the event of an immediate emergency, call the UNL Police at 402-472-2222.

Title IX

UNL prohibits discrimination and harassment based on sex, including sexual misconduct, stalking, and dating violence.

• Reporting: All University employees (unless designated as confidential) are required to report incidents of sexual misconduct to the Title IX Coordinator.

• Resources: For more information on your rights and available resources, visit the Office of Institutional Equity and Compliance website.

Student Support Resources

• School of Computing Student Resource Center: Located in Avery 13A (cse.unl.edu/src), where you can go for assistance with any of your computing courses.

• University Writing Center: Located in the local campus hubs, providing trained peer consultants to help you plan, draft, and revise your writing. Visit www.unl.edu/writing/ for hours and appointments.

• Central Repository: You can view a comprehensive list of university-wide requirements at go.unl.edu/course policies.

Copyright Information and Project Release

All course materials, including slide presentations, digital assets, and digital twins, are protected by copyright law and University of Nebraska-Lincoln policy. As the instructor, I am the exclusive owner of the copyright for these original materials, which represent years of my research and development. Students currently enrolled in this course may take notes, make copies of course materials, and utilize the provided assets solely for their own personal learning, and they may share these materials with fellow classmates. However, you may not share these materials online, make them open-source, or reproduce, distribute, display, or upload lecture notes, recordings, or course assets in any other way without my express prior written consent. Allowing others to do so is also strictly prohibited and may subject you to student conduct proceedings under the UNL Student Code of Conduct. To acknowledge these terms, every student must submit the Course Statement of Understanding before the end of the first week of the semester.

To experience AR development over these 16 weeks, I am sharing my proprietary scripts, repositories, 3D models, and other digital assets. For each assignment, you will develop on top of foundational repository, which must be set to private and include correct citations for any external code or assets. All project submissions will proceed exclusively through GitHub, and failure to meet these configuration or citation requirements will result in a point deduction.

Because these course materials were created as functional research tools, using or distributing any research or assets outside of this class requires prior approval from the instructor, You-Jin Kim. This prior approval is mandatory for activities including, but not limited to, academic or commercial competitions, graduate school applications, and portfolio displays. Furthermore, if you plan to publish a scientific article or journal paper based on work from this course, you must obtain a formal project release. The research team behind these materials must be recognized for their contribution and properly acknowledged in the publication. As explicitly stated in the Course Statement of Understanding, failure to receive proper permission before sharing or publishing these materials externally violates ethical standards in the research community, breaches university policy, and is subject to further action.

© You-Jin Kim
        Nebraska–Lincoln 🌽