Unlock High-Quality Game Character Art with This Cutting-Edge Workflow
1. Workflow Overview
This workflow generates high-quality orthographic views (front/side/back/close-up) for game characters, ideal for concept art. Key features:
Flux model architecture for stable generation.
ControlNet ensures character consistency across angles.
Character-design LoRA enhances details.
2. Core Models
Model Name | Function | Source |
|---|---|---|
flux1-dev-fp8.safetensors | Base model (FP8 optimized) | Download from HuggingFace |
CharacterDesign-FluxV2 | LoRA for character details (weight=0.8) | Requires plugin installation |
FLUX.1-Union-Pro-Shakker | ControlNet model for multi-angle consistency | Separate download required |
Dual Text Encoders |
| Auto-downloaded |
3. Key Nodes
Node Name | Function | Installation | Dependencies |
|---|---|---|---|
ControlNetApplyAdvanced | Applies ControlNet (strength=0.7) | Built-in |
|
CLIPTextEncodeFlux | Flux-specific text encoder | Install | T5XXL + CLIP-L models |
LoraLoaderModelOnly | Loads character-design LoRA | Built-in |
|
Primitive string multiline | Dynamic prompt generation | Install | None |
4. Workflow Structure
Group 1: Model & Prompt Loading
DualCLIPLoader: Loads text encoders.
Primitive string multiline: Generates prompts (e.g., "elf princess in blue dress").
Group 2: Generation Control
ControlNetLoader: Loads FLUX ControlNet.
ControlNetApplyAdvanced: Constrains poses (uses OpenPose reference).
Group 3: Output
KSampler: 20 steps, Euler, CFG=1.
VAEDecode: Renders 2048x1024 HD images.
5. Inputs & Outputs
Input Parameters:
Required: Character description, OpenPose reference image.
Optional: Seed (random), LoRA strength (default=0.8).
Output: Multi-angle PNGs (clean background, no logos).
6. Notes
Compatibility: Must use
fp8_e4m3fnprecision models.VRAM: ≥12GB recommended; batch processing for 2048x1024 resolution.
Plugins: Install
FluxandCrystoolsvia ComfyUI Manager.Debug: Verify OpenPose input if poses are incorrect.