Practical overview
When it fits
Use Piti when the desired image-generation input is a semantic map or sketch and a local PyTorch workflow, an interactive Gradio interface, or a batch script fits the task.
Who it is for
- Visual designers and concept artists exploring image variations from masks or sketches.
- Researchers and developers working on image-to-image synthesis who want inference and model fine-tuning scripts.
Documented uses
- Generate images from RGB semantic maps.
- Generate images from sketch maps extracted with HED.
- Run interactive or batch image inference with pretrained checkpoints.
- Fine-tune the Base and Upsample models for mask-to-image or sketch-to-image tasks.
Requirements & limits
The repository is public and licensed under MIT. The README also links to an online demo; local use requires repository setup and pretrained checkpoints.
- The documented pretrained tasks are mask-to-image and sketch-to-image, with separate checkpoint pairs for each.
- Mask inputs use RGB semantic maps; sketch inputs use maps extracted by HED.
- Local inference requires both Base and Upsample checkpoints, and batch inference uses a text file of input image paths.
- The project provides preprocessing scripts, interactive Gradio inference, batch inference and fine-tuning scripts.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run interactive mask- or sketch-to-image inference
Prepare the documented local environment and launch Piti's interactive inference interface.
Before you start
- A local Python environment with Conda.
- The Piti repository and its environment.yml file.
- A pretrained Base and Upsample checkpoint for the chosen task, placed in ./ckpt.
- An input semantic map for mask-to-image or sketch map for sketch-to-image.
Steps
Clone the repository and change into its directory.
Create the environment with conda env create -f environment.yml.
Download the appropriate pretrained Base and Upsample models and place them in ./ckpt.
Install Gradio with pip install gradio, then run python inference.py and upload a mask or sketch.
Expected outcome
An interactive Gradio interface for generating images from uploaded masks or sketches.
Guide limitations
- The README describes the input types but does not specify hardware requirements.
- The listed checkpoints are task-specific; choose the mask-to-image or sketch-to-image models that match the input.
Generate images in batch from prepared inputs
Configure the documented batch inference script for a selected input type and pretrained models.
Before you start
- The repository environment and relevant Base and Upsample checkpoints.
- A text file containing paths for the images used as inputs.
Steps
Modify sample.sh to set the base model path, upsample model path, input-path list, number of samples, guidance strength and input mode.
Run bash sample.sh.
Expected outcome
The batch inference script is configured to sample images for the selected input type.
Guide limitations
- The README does not provide a detailed output-location description.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- PITI-Synthesis/PITI · Official homepage title · retained in its source language