Where We Are

• Part 1: skeleton hierarchy, local controls, world-space FK
• Part 2: rest pose, skinning weights, one-hot binding, LBS
• Part 3: motion clips, avatars, scene timing, camera, final evidence

Part 3: Build A 30s+ Avatar Animation Scene

Work in groups of two to create a short animated scene featuring virtual human characters.

• at least 30 seconds long
• coherent scene, not isolated test clips
• final video, saved scene files, and still frames for the report
• technical explanation of your workflow and limitations

Scene Motion Editor

The GF5 scene motion editor is the main Part 3 composition tool.

Reference: Scene Editor Manual

What The Editor Lets You Do

• preview bundled motion presets
• arrange clips on character tracks
• edit Path, facing, timing, and camera
• save .scene.json drafts
• export screenshots or video evidence

Editor Demo

• Motion Library: choose preset or custom clips
• Timeline: arrange clips per character track
• Stage: edit root path and facing
• Final avatar: assign proxy, SMPL, or extracted avatar package
• Export: save draft evidence

A Simple Way To Work

1. Build a rough scene with the provided motion library.
2. Preview it on the SMPL-24 Proxy, then adjust timing, Path, and facing.
3. Save the scene as a .scene.json file.
4. Try custom avatars, generated motions, or other tools as needed.
5. Render the final video and improve the result as much as you can.

Motion, Path, Facing, Camera

Path controls the character root trajectory.

• motion clip: local body pose over time
• Path: where the body root travels in the scene
• facing: which way the character looks while moving
• camera: how the audience reads the staged action

Motion Library

Preset clips are grouped by intended use.

• Standing / Gesture: mostly stay on one spot
• Travel Loops: repeated locomotion along a path
• Travel Transitions: starts, stops, and side steps
• Turns: changes in facing direction
• Special Actions: larger spot actions or original-travel beats

Good Scene Ingredients

• a small number of readable actions
• clear entrances, reactions, or final poses
• enough holds for the viewer to understand the action
• simple camera framing
• visible comparisons between draft and final

Custom Avatars

The GF5 UP2You demo can reconstruct a custom virtual character for final rendering.

• download the avatar package
• extract it under libraries/avatars/
• refresh the scene editor
• choose it from Final avatar
• compare proxy preview against final avatar render

Avatar Package Layout

libraries/avatars/alex/outputs/animation_lowres.obj
libraries/avatars/alex/outputs/animation_lowres_skinning_weights.npz
libraries/avatars/alex/outputs/smplx_mesh.obj

Extract each downloaded ZIP into its own named folder under libraries/avatars/.

One Skeleton Connects The Pieces

GF5 uses the same 24-joint body skeleton across Part 3.

• proxy character
• preset motion clips
• imported or generated clips
• skinning weights
• custom avatar package

SMPL-X To SMPL-24

Some reconstruction tools use SMPL-X internally.

• UP2You fits an SMPL-X-style body while reconstructing the avatar
• GF5 exports weights back to the course 24-joint skeleton
• the final avatar can follow the same clips as the SMPL-24 Proxy

Text-to-Motion Generation

HY-Motion-1.0 can generate short candidate motions from text prompts.

Open the HY-Motion-1.0 demo

Using Generated Motion Clips

• use one physical action per prompt
• keep the downloaded .fbx and .txt prompt together
• import the pair into the Motion Library
• treat generated clips as ingredients, not the whole scene

Prompt Atomic Actions

Good prompts describe one physical action.

• A person walks forward and stops.
• A person turns left and waves.
• A person points forward with the right hand.
• A person performs a short dance step.
• A person stands still and raises both arms.

Import HY-Motion Clips

Keep the .fbx animation and matching .txt prompt together.

wave_to_camera.fbx
wave_to_camera.txt

Use Import HY-Motion in the Motion Library panel and select both files.

The Scene Editor auto-finds motion clips in:

libraries/motions/custom/

Imported clips are saved there. After import, refresh the Scene Editor page and load the clip from Custom:.

Hand Crafting Motion

You could also hand craft motions in the Asset Viewer.

Asset Viewer motions are saved to the same custom motion folder.

After saving a motion, refresh the Scene Editor page and load it from Custom:.

Motion clips from Parts 1 and 2 work too; move them into this folder if they are somewhere else (recently changed from libraries/motions/saved/ to libraries/motions/custom/).

Week 3

Build a proof of concept

• explore the bundled motion library
• make a rough scene on the SMPL-24 Proxy
• reconstruct or choose character assets
• save one or more .scene.json drafts
• bring a screenshot or clip to the help session

Week 4

Refine and submit

• improve timing, staging, and camera
• compare proxy preview against final avatar render
• collect final stills, clips, and artifact examples
• export a 30-second (or longer) final video
• write the final report

Artifacts To Look For

• foot sliding
• scale or orientation mismatch
• skinning collapse near joints
• clothing or appearance loss
• awkward transitions between clips
• camera framing that hides the action

Improve Rendering Quality

• Should we adjust the camera trajectory or framing?
• Is it possible to add a background scene or simple environment?
• How can the motion or avatar look more realistic?

Interim Report Feedback

Brief feedback on your interim reports has been uploaded to Moodle.

• you all did a very good job
• the tasks may have been a little too easy
• grades were spread a bit by report quality and the motion sequence created

A Few Report Notes

• start with a brief summary of the report
• write equations using proper mathematical notation
• make FK robust to unsorted joint lists

FK Ordering Pitfall

FK computes each joint world transform from its parent:

W[j] = W[parent[j]] @ L[j]

• this requires the parent's W to be ready first
• a plain loop only works if parents appear before children
• counterexample order: wrist, elbow, shoulder
• robust FK should sort parent-to-child, or recurse

Part 3 Showcase

Tuesday 9 June 2026, 11am-1pm, LT6.

Showcase notes

• 9 groups
• 5 minutes presenting + 2 minutes Q&A per group
• 10 marks, so keep it light
• we will connect group laptops
• start with the animation, then tell the quick story
• share one trick, surprise, or useful failure

Presentation Order

Order	Group	Order	Group
1	Matthew & Encheng	6	Elen & Sarah
2	Madeleine & Kavita	7	Yilia & Frank
3	Rohan & Sam	8	Tianze & Yunge
4	Serene & Aaron	9	Donna & Stan
5	Zohaib & Sean

Final Report

Deadline: Friday 12 June 2026, 4pm.

• Final Report requirements
• Submit on Moodle
• 8 pages max: up to 6 pages group work + up to 2 pages individual
• explain the scene and pipeline
• include final result plus a few useful figures
• discuss artifacts, limitations, and what you learned
• include the AI Use Statement

Report Structure

• Scene Plot
• Method
• Results And Evidence
• Artifacts, Limitations, And Reflection
• AI Use Statement
• Individual Contribution And Insights

AI Use In Part 3

• AI tools may support creative and production work
• generated motion, assets, editing, coding help, or writing support must be declared
• reports must still accurately explain your own workflow and evidence
• include an AI Use Statement

What Success Looks Like

• a readable 30-second (or longer) scene
• motion choices that match the intended action
• clear staging and camera
• a final render that improves on the draft
• honest discussion of artifacts and limitations
• reproducible evidence of the workflow

What To Bring To Help Sessions

• show the editor, a screenshot, or a short clip
• point to the part that feels wrong or unfinished
• bring one concrete question, idea, or experiment you want to try

Where To Go Next

• Part 3 handout
• Scene editor manual
• Showcase notes
• Final report