Deep Learning Model Architectural Diagrams -- 2

I need high-quality architecture diagrams for **three deep learning models** used in trajectory prediction (diffusion-based). The style should be clean, academic, and consistent across all diagrams (similar to the attached reference).

---

## General Requirements

* Style: minimal, vector-based (paper-quality figure)
* Format: PDF + editable source (Figma / Illustrator / PowerPoint)
* All diagrams must share the SAME layout for easy comparison
* Each diagram has 3 stages:

1. Noising
2. Embedding / Conditioning
3. Denoising (attention-based network)


## Model 2: Additive Conditioning

* Conditioning features are added to embeddings
* Show:

* Separate condition tensor
* Merge using "+" before entering the network
* Conditioning happens BEFORE attention layers

---

## Model 3: Conditioning Inside Attention Layers

* Conditioning is injected INTO attention modules:

* Agent-to-Agent attention (condition-aware)
* Agent-to-Roadgraph attention (condition-aware)
* Show arrows from condition → attention blocks
* NOT added to embeddings

---

## Model 4: FiLM Conditioning (Feature-wise Linear Modulation)

* Conditioning is applied via **FiLM layers inside the network**
* FiLM applies:

* Scaling (γ) and shifting (β) to features

### Important:

* Show conditioning → small module → outputs (γ, β)
* Then (γ, β) modulate intermediate features inside the network

### Visually:

* Add a small block labeled "FiLM (γ, β)"

* Show it modifying feature maps inside layers:

* e.g. Feature → (γ ⊙ Feature + β)

* This should be shown inside the denoising network (not at i