CPU3DAI tools for 3D, video and audio

Allegro: video generator — what it does and what you need to run it

Allegro is an open source video generator from Rhymes AI that tackles the text-to-video task. The model turns a text description into a short clip and is distributed freely: both the code and the weights are released under the Apache-2.0 license.

What it does

As described by the author, Allegro generates videos up to 6 seconds long at 15 frames per second and 720p resolution. The model is built on the diffusers library and takes a simple text description as input. The authors do not specify which prompt languages are supported, but the examples in the repository are in English.

What you need to run it

Allegro is written in Python. You will need PyTorch version 2.4 or higher and CUDA 12.4 or newer — the authors explicitly mention only the CUDA platform. The full list of dependencies is in requirements.txt.

The largest model weights file is 8.9 GB, and the full set is 41.3 GB. As described by the author, without special measures the model requires about 27.5 GB of VRAM. If you enable --enable_cpu_offload, usage drops to roughly 9.3 GB, but inference time increases noticeably. The minimum threshold the developers mention in the context of memory savings is 3 GB, though this figure is not elaborated on verbatim in the documentation.

The code is available in the repository github.com/rhymes-ai/Allegro, and the weights are on huggingface.co/rhymes-ai/Allegro. Project website: rhymes.ai.

Who it is for

Allegro is aimed at those looking for an open source video generator that can be run on your own hardware. The model suits researchers and developers willing to deal with setting up a CUDA environment and managing memory via CPU offload. If you have a GPU with less than 9–10 GB of VRAM, you should count on the CPU offload mode and accept slower performance.

Allegro is not suitable for those who need generation longer than 6 seconds, a frame rate above 15 FPS, or operation without an NVIDIA GPU with CUDA support. The model is also not designed for users expecting a ready-made application with an interface: it runs via Python scripts.

Allegro is one of the few fully open source video generators where both the code and the weights are available under the permissive Apache-2.0 license. The repository was created in October 2024 and updated in early February 2025, so the project is in an active development phase.

Allegro pipeline — text-to-video Rhymes AI · open source · Apache-2.0 Text prompt simple description in natural language Tokenization splitting into tokens text embeddings Diffusion model frame generation noise → 720p video 15 FPS, up to 6 sec VRAM: 9.3–27.5 GB Video assembly frame stitching encoding Final video 720p, 15 FPS, up to 6 sec format: MP4 Requirements Python ≥ 3.10 PyTorch ≥ 2.4 CUDA ≥ 12.4 Library: diffusers CPU offload
How the Allegro pipeline works. The diagram is drawn from the tool’s fact sheet.

Fact sheet

Tasktext-to-video source
Code licenseApache-2.0 source
Weights licenseapache-2.0 source
PlatformCUDA per the author’s description source
VRAM9.3–27.5 GB per the author’s description source
Largest weights file8.9 GB source
All weights files41.3 GB source
Librarydiffusers source
LanguagePython source
Last code change2025-02-07 source
Repository created2024-10-16 source
Changes often — as of 2026-08-19
GitHub stars1132 source
Forks69 source
Downloads per month109 source
Model updated2024-10-31 source

Values are collected automatically from official sources and were checked on 2026-08-19. Each one links to its source, and values taken from the developer’s pages also carry a verbatim quote — hover over the note. Pricing and versions are shown as of the check date and change most often; verify on the vendor’s site before buying.

See also