T2I-Interp Toolkit

A text-to-image interpretability toolkit for steering, SAE analysis, stitching, and cross-attention localisation in diffusion models.

Features

• Activation steering over UNet modules
• Sparse autoencoder (SAE) analysis workflows
• Latent stitching across layers
• Cross-attention localisation sweeps
• Hydra-driven config and multirun support

Installation

# from repository root
pip install -e .

Optional auth for datasets/experiment tracking:

huggingface-cli login
wandb login

CLI Quickstart

Both command styles are supported:

t2i steer
t2i-steer

Primary workflows:

# Steering
t2i-steer
t2i-steer prompt="a cinematic shot of a happy professor" refresh_batch_size=64
t2i-steer -m layer_names="[unet.down_blocks.1.attentions.0.transformer_blocks.0.attn2,unet.mid_block.attentions.0.transformer_blocks.0.attn2]"

# Stitch
t2i-stitch
t2i-stitch prompt="A red car turning into a blue car"

# SAE
t2i-sae
t2i-sae n_top_features=6 num_inference_steps=2

# Localisation
t2i-localise
t2i-localise -m guidance_scale=0.0,2.0,5.0

W&B override example:

t2i-localise wandb.project="attention-ablation" wandb.name="baseline-sweep"

Config Locations

• t2i_interp/config/steer/run.yaml
• t2i_interp/config/stitch/run.yaml
• t2i_interp/config/sae/run.yaml
• t2i_interp/config/localisation/run.yaml
• t2i_interp/config/wandb.yaml

Notebooks

• notebooks/steer.ipynb
• notebooks/stitch.ipynb
• notebooks/sae.ipynb
• notebooks/localisation.ipynb

Project Structure

T2I_Interp_toolkit/
├── t2i_interp/
│   ├── cli.py
│   ├── config/
│   │   ├── steer/
│   │   ├── stitch/
│   │   ├── sae/
│   │   └── localisation/
│   ├── scripts/
│   │   ├── run_steer.py
│   │   ├── run_stitch.py
│   │   ├── run_sae.py
│   │   └── run_localisation.py
│   └── utils/
├── bash/
├── notebooks/
└── pyproject.toml

Development

git clone https://github.com/Social-AI-Studio/T2I_Interp_toolkit.git
cd T2I_Interp_toolkit
pip install -e ".[dev]"

Before opening a PR:

ruff check .
ruff format .
pytest

License

MIT