Assignments

📝 Assignment 1: The Creative Brain & Multi-View Ideation (2D Pipeline)

Agency Phase Replaced: The Concept Art & Ideation Department.

Objective: Establish the automated foundation for design logic and multi-view generation.

The Task:

Use DeepSeek (R1 or V3) to act as your "Design Director." Write a system prompt that forces DeepSeek to output highly structured, engineering-focused design descriptions and optimized image-generation prompts for a complex object (e.g., futuristic eyewear, ergonomic tools, or sci-fi vehicle parts).
Pipe DeepSeek’s output into an open-source image model like FLUX.1 or Stable Diffusion XL inside ComfyUI or via a local Python script.
The goal is to consistently generate an accurate, high-fidelity multi-view turnaround sheet (front, side, back views) of the object against a solid neutral background.

Deliverable: A functional script or ComfyUI workflow file that takes a simple product concept text prompt and automatically exports a clean, usable multi-view design sheet.

Grading Check Point and Rubric:

Objective: Build an automated text-to-concept turnaround pipeline.

[ ] 1. System Prompt Optimization: Design a robust system prompt forcing DeepSeek (or Meta Llama) to act as a highly specialized industrial design director.
[ ] 2. Structured Output Parsing: Configure the LLM to output clean, programmatically parsable data structures (like JSON) containing design constraints, materials, and automated image-generation prompts.
[ ] 3. Prompt Engineering for Continuity: Embed explicit tokens in the image prompts (e.g., "orthographic turnaround," "isometric view,") to minimize artistic flair and maximize structural utility.
[ ] 4. Model Pipeline Setup: Successfully deploy an open-weights image model—such as Black Forest Labs FLUX.1 or Stable Diffusion XL—within a local environment or ComfyUI node network.
[ ] 5. View Consistency Control: Implement steering mechanics (ControlNet or IP-Adapter) to ensure the generated front, profile, and back views share identical design languages, colors, and proportions.
[ ] 6. Background Isolation: Force a perfectly clean, solid alpha-ready background (white, gray, or transparent) to prevent environment clipping during the 3D phase.
[ ] 7. Aspect Ratio & Resolution Calibration: Lock the visual outputs to a minimum resolution of $1024 \times 1024$ pixels per sheet to guarantee pixel density for spatial extraction.
[ ] 8. Automated Parameter Scripting: Script the batching mechanism so a single input concept automatically yields three highly detailed, distinct design variations without manual rebuilding.
[ ] 9. Engineering Documentation: Provide a written specification sheet generated by the LLM detailing the structural logic, ergonomics, and material properties of the concept.
[ ] 10. Workflow Serialization: Export and submit the finalized pipeline configuration file (.json or ComfyUI workflow) to prove local pipeline reproducibility.

📦 Assignment 2: 2D-to-3D Surface Geometry Extraction

Agency Phase Replaced: The 3D Digital Modeler / Digital Sculptor.

Objective: Convert flat 2D concept sheets into functional 3D digital geometries using local hardware and validate spatial bounds in Blender or Unity.

The Task:

Take your optimized multi-view images from Assignment 1 and pass them into a cutting-edge open-source 3D generative model such as Microsoft TRELLIS 2 or Tencent Hunyuan3D.
Experiment with the model parameters to generate an explicit 3D mesh.
Analyze the resulting topology, vertex count, and texture projection mapping. Identify where the AI "hallucinates" geometry on hidden angles.

Deliverable: A raw .obj or .glb mesh generated entirely locally, paired with a short technical analysis explaining the geometric accuracy and topological flaws of the model.

Grading Check Point and Rubric:

Objective: Transform flat 2D turnaround concept sheets into functional digital meshes.

[ ] 1. Image Pre-processing Automated Pipeline: Implement automated background removal or alpha-channel masking to isolate the object before it passes to the 3D generator.
[ ] 2. 3D Model Local Deployment: Successfully instantiate a local or API instance of an advanced open-source 3D generator like Tencent Hunyuan3D or Microsoft TRELLIS.
[ ] 3. Inference Parameter Tuning: Experiment with and document optimal generation settings (e.g., adjustment of steps, guidance scale, and conditioning views) to balance rendering speed against accuracy.
[ ] 4. Mesh File Generation: Successfully export raw, explicit 3D geometry assets in standard production formats (.obj, .glb, or .stl).
[ ] 5. Topology & Vertex Analysis: Conduct an architectural review of the generated mesh topology, documenting polygon counts, edge loops, and structural anomalies.
[ ] 6. Texture Mapping & UV Inspection: Evaluate the continuity of baked PBR textures or vertex colors across complex geometries, identifying any visual stretching or alignment issues.
[ ] 7. Hallucination Mapping: Identify, screenshot, and technically evaluate where the AI "hallucinated" geometric forms in occluded zones (blind spots between the 2D input angles).
[ ] 8. Dimensional Calibration: Explicitly calibrate the digital bounding box units within Unity or Blender to map perfectly to real-world millimeters or inches.
[ ] 9. Ground-Truth Deviation Test: Graphically compare the silhouette of the generated 3D mesh against the original 2D turnaround sheets to measure geometric error.
[ ] 10. Technical Geometry Report: Submit a concise technical analysis detailing the performance, structural limits, and topological breakdown of the 3D generation model.

🖨️ Assignment 3: Mesh Optimization and Additive Manufacturing

Agency Phase Replaced: The Technical Designer & Rapid Prototyping Technician.

Objective: Bridge the gap between fragile "AI-generated digital shells" and real-world manufacturability.

The Task:

Raw AI meshes are notorious for being non-manifold (having holes, floating geometry, or overlapping faces). In this phase, you must create a remediation workflow.
Use open-source tools (like Blender's Python API or automated script tools) to repair the mesh, decimate unnecessary polygons, and ensure the model is completely "watertight."
Import your repaired asset into an open-source slicer (like OrcaSlicer or Cura), configure your print settings, and run a physical test-print of your object.

Deliverable: A successful, physical 3D-printed draft of your AI-generated object, proving your digital pipeline creates tangible, real-world assets.

Grading Check Point and Rubric:

Objective: Clean unstable AI geometries and successfully fabricate a physical test draft.

[ ] 1. Non-Manifold Repair Execution: Use automated scripting tools (or Blender's 3D Print Toolbox) to locate and repair open edges, non-manifold geometry, and self-intersections native to AI generation.
[ ] 2. Watertight Mesh Validation: Prove the asset is 100% "watertight" by running a mesh-integrity algorithm showing zero geometric holes or floating vertices.
[ ] 3. Polygon Decimation Optimization: Apply controlled polygon reduction algorithms to reduce file sizes for manufacturing pipelines while preserving critical design features.
[ ] 4. Wall Thickness Calculations: Perform a cross-sectional thickness evaluation to ensure no walls are thinner than the structural printing limit (minimum 1.6–2.0 mm recommendation).
[ ] 5. Slicer Software Integration: Successfully import the repaired file into an open-source slicing engine (OrcaSlicer, PrusaSlicer, or Cura) with clear profile documentation.
[ ] 6. Orientational Efficiency Optimization: Programmatically or manually calculate the optimal printing orientation to maximize surface quality and minimize structural print failures.
[ ] 7. Support Structure Management: Configure appropriate support structures (tree or grid patterns) designed to prevent overhang collapse on complex, AI-generated contours.
[ ] 8. Infill & Slicing Architecture Setup: Select and defend your structural choices for infill density (minimum 15%), infill pattern (e.g., Gyroid for omnidirectional strength), and layer height.
[ ] 9. Successful 3D Draft Print: Execute a complete physical print of the object using standard prototyping materials (PLA, PETG, or resin) with zero terminal print failures.
[ ] 10. Physical Post-Mortem Assessment: Provide a side-by-side photographic comparison of the physical prototype and the digital model, identifying real-world material failures caused by AI geometric.

Page updated

Google Sites

Report abuse