Practical overview
When it fits
Use Llamagen when you want model weights and research code for autoregressive image generation, rather than a packaged image-editing application. Choose the text-conditional workflow for prompt-driven images and the class-conditional workflow for generation by class.
Who it is for
- Researchers studying autoregressive image generation and its relationship to diffusion methods.
- Developers who want pretrained checkpoints and code as a basis for text-to-image or class-conditional generation.
- Teams exploring training or serving workflows for autoregressive image models.
Documented uses
- Generate images from text prompts using released text-conditional models.
- Generate class-conditional images using the provided sampling scripts.
- Use released PyTorch code and model weights as a starting point for training and evaluation.
- Explore serving the released models with the vLLM framework identified by the project.
Requirements & limits
The repository is public and reports an MIT license for most of the project; some portions are covered by separate licenses. The README links to an online demo as well as local workflows.
- The project offers both text-conditional and class-conditional generation, but they use different checkpoints and sampling scripts.
- The README specifies PyTorch 2.1 or later for training and sampling code.
- Text-conditional sampling additionally requires packages and language models covered in language/README.md.
- Class-conditional model sizes span roughly 100 million to 3.1 billion parameters, which may affect checkpoint selection and resource needs.
- The README identifies vLLM support for serving and links to an online Hugging Face demo.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Sample a class-conditional image locally
Use a listed pretrained class-conditional checkpoint and sampling script to generate an image file.
Before you start
- Python 3 and PyTorch 2.1 or later.
- Downloaded class-conditional tokenizer and model checkpoints placed in ./pretrained_models.
- The repository's sampling code and its dependencies.
Steps
Download a class-conditional tokenizer checkpoint and an AR model checkpoint listed in the repository.
Place both files in ./pretrained_models.
Run the documented sample_c2i.py command with the matching checkpoint paths and model settings.
Expected outcome
The script saves the generated image as sample_c2i.png.
Guide limitations
- The README provides example settings for 384-pixel generation; it does not specify a general hardware requirement.
- The FSDP example uses an additional --from-fsdp flag.
Sample a text-conditional image locally
Run one of the documented text-to-image sampling workflows with pretrained checkpoints.
Before you start
- Python 3 and PyTorch 2.1 or later.
- The required packages and language models described in language/README.md.
- Downloaded text-conditional tokenizer and AR checkpoints placed in ./pretrained_models.
Steps
Follow language/README.md to install the required packages and language models.
Download the documented text-conditional tokenizer and either the stage-one or stage-two AR checkpoint.
Place the checkpoints in ./pretrained_models.
Run the corresponding sample_t2i.py command from the repository README.
Expected outcome
The script saves the generated image as sample_t2i.png.
Guide limitations
- The documented examples target 256- or 512-pixel output.
- The README does not specify a general hardware requirement.
Compare options for Model inference and serving
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Llamagen | Generate images from text prompts using released text-conditional models. | Linux, Server | The repository is public and reports an MIT license for most of the project; some portions are covered by separate licenses. The README links to an online demo as well as local workflows. |
| FASHN VTON v1.5 | Generate a try-on preview from a person image and a garment image. | CLI, Linux, macOS, Self-hosted | The repository is public and states an Apache-2.0 license. Local setup requires downloading approximately 2 GB of model weights. |
| Posteromni | Rescale a poster to a target aspect ratio using a text prompt. |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- FoundationVision/LlamaGen · Official homepage title · retained in its source language