IC-Custom Model Card
August 31, 2025 · View on GitHub
This document provides a comprehensive overview of all models used in IC-Custom.
🎯 IC-Custom Models
Our core customization models, automatically downloaded when running the app:
| Model | Purpose | Source |
|---|---|---|
| dit_lora | LoRA adaptation for diffusion transformer | TencentARC/IC-Custom |
| img_txt_in | Image-text input layers weights | TencentARC/IC-Custom |
| boundary_embeddings | Boundary condition embeddings | TencentARC/IC-Custom |
| task_register_embeddings | Task registration embeddings | TencentARC/IC-Custom |
🔧 Base Models
Foundation models required for IC-Custom operation:
| Model | Purpose | Source |
|---|---|---|
| CLIP | Vision-language understanding | openai/clip-vit-large-patch14 |
| T5 | Text processing | google/t5-v1_1-xxl |
| SigLIP | Image understanding | google/siglip-so400m-patch14-384 |
| Autoencoder | Image encoding/decoding | black-forest-labs/FLUX.1-Fill-dev |
| DIT | Diffusion model | black-forest-labs/FLUX.1-Fill-dev |
| Redux | Image processing | black-forest-labs/FLUX.1-Redux-dev |
🎨 App Interactive Models
Additional models for enhanced app functionality:
Required for App
| Model | Purpose | Source |
|---|---|---|
| SAM-vit-h | Image segmentation | HCMUE-Research/SAM-vit-h |
Optional (Enable via CLI flags)
| Model | Purpose | Source | Enable Flag |
|---|---|---|---|
| BEN2 | Background removal | PramaLLC/BEN2 | --enable_ben2_for_mask_ref True |
| Qwen2.5-VL | Prompt generation | Qwen/Qwen2.5-VL-7B-Instruct | --enable_vlm_for_prompt True |
🔗 Model Usage
- Automatic Download: All required models are downloaded automatically when running scripts
- Manual Configuration: Model paths can be specified in
app.yamlorinference.yaml - Optional Models: BEN2 and Qwen2.5-VL are disabled by default and only downloaded when explicitly enabled
For detailed configuration instructions, see: