OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Time Diffusion

Time Diffusion is a Python research implementation of TIME, a method for editing implicit assumptions in text-to-image diffusion models. It includes experiment scripts that apply editing functions to prompt datasets, generate images and save results, with options for textual augmentations and editing key as well as value projections.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Time Diffusion — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use the repository when you want a command-line research workflow for applying TIME across a dataset or to inspect the linked browser demo for interactive editing.

Who it is for

  • Researchers studying text-to-image diffusion model editing.
  • Developers comparing editing functions or running prompt-dataset experiments.

Documented uses

  • Apply a model-editing method to prompts in a dataset.
  • Compare the closed-form editing function with a baseline.
  • Generate multiple image results per prompt and save them for analysis.
  • Explore model edits through the linked browser demo.
Before you choose

Requirements & limits

  • The experiment script supports choosing train_closed_form or baseline; the regularization parameter is used with the closed-form function.
  • --with_augs enables textual augmentations, while --with_to_k includes editing the key projection along with the value projection.
  • Runs select a dataset CSV, inclusive starting index, exclusive ending index, and number of seeds per prompt.
  • The repository reports a specific tested software stack; other runtimes and model configurations are experimental.
  • A Gradio browser demo is linked, but the repository does not establish whether it is maintained or deployed by the repository.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run a dataset-based TIME experiment

Apply the documented editing function across a selected prompt range and save generated image results.

Before you start

  • Python environment based on the repository's environment.yml example.
  • The reported test stack is PyTorch 1.13.1, CUDA 11.6.2, NumPy 1.23.4 and Diffusers 0.9.0.
  • A dataset CSV; the README identifies TIMED_test_set_filtered_SD14.csv as the default.

Steps

  1. Choose train_closed_form or baseline for --train_func; set --lamb when using the closed-form function.

  2. Set --dataset, --begin_idx, --end_idx and --num_seeds to select prompts and image seeds.

  3. Run the command below with the selected options, optionally adding --with_to_k and --with_augs; set --save_dir for the output directory.

    Documented command
    python apply_time.py

Expected outcome

The experiment applies TIME over the selected dataset range and saves generated results in the specified directory.

Guide limitations

  • The repository reports testing only with PyTorch 1.13.1, CUDA 11.6.2, NumPy 1.23.4 and Diffusers 0.9.0; other configurations are not established.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
24 October 2026
Original website title
bahjat-kawar/time-diffusion · Official homepage title · retained in its source language
How we review sourcesSuggest a correction