Practical overview
When it fits
Use Liquid when you want to experiment with text and image tasks through the repository's inference scripts or local Gradio demo.
Who it is for
- Researchers studying unified multimodal language models.
- Developers exploring autoregressive image generation and image understanding.
Documented uses
- Generate text responses from a prompt.
- Ask questions about an image.
- Generate images from text prompts.
- Explore a model architecture that unifies visual comprehension and generation.
Requirements & limits
The public repository lists an MIT license. The repository provides instructions for local inference and a local Gradio demo.
- Inference uses a Hugging Face format model and the Transformers library.
- The repository provides distinct scripts for text dialogue, image understanding and text-to-image generation.
- Image generation on GPUs with less than 30GB VRAM may require 8-bit loading.
- Training and data-processing instructions are in separate repository documents.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run the local Gradio demo
Start the repository's Gradio demo locally.
Before you start
- Python environment with the repository available
- Install gradio==4.44.1 and gradio_client==1.3.0
Steps
Install the listed Gradio and Gradio Client versions.
Change to the evaluation directory.
Run python app.py.
Expected outcome
The repository's Gradio demo starts locally.
Guide limitations
- For image generation on a GPU with less than 30GB VRAM, the repository says 8-bit loading may be needed to avoid out-of-memory errors.
- The README does not specify broader hardware requirements.
Run a single text, image-understanding or image-generation inference
Use the repository's supplied single-inference scripts with the Liquid_V1_7B model.
Before you start
- The repository and its inference dependencies, including Transformers
- Liquid_V1_7B model path Junfeng5/Liquid_V1_7B
- An image path for image-understanding inference
Steps
For text dialogue, use inference_t2t.py with the model path and a prompt.
For image understanding, use inference_i2t.py with the model path, image path and prompt.
For text-to-image generation, use inference_t2i.py with the model path and prompt.
Expected outcome
The selected script performs text dialogue, image question answering or image generation.
Guide limitations
- The repository recommends consulting evaluation/EVAL.md for dependency versions.
- For image generation on a GPU with less than 30GB VRAM, the README recommends adding --load_8bit.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- FoundationVision/Liquid · Official homepage title · retained in its source language