How to run OmniGen2 locally

BAAI / VectorSpaceLab · Any to image (gen + edit + in context) · 7B params · Apache-2.0 (commercial OK)

Dual-path model (a 3B Qwen2.5-VL understanding head plus a 4B diffusion head) that does text-to-image, instruction-based editing and in-context generation in one checkpoint with state-of-the-art open editing quality.

What it costs to run — $0 markup

On your own machine — $0. Runs free if you have about 17 GB of VRAM (~17GB on a single RTX 3090/4090 natively; model CPU-offload roughly halves it, sequential offload drops under 3GB at a large speed cost.), via Diffusers.
Rent a GPU — from $0.26/hr. Fits RTX 3090 24GB at the direct vendor price ($0 markup) — pay only for the minutes you generate.
Download the weights — free. Open weights at OmniGen2/OmniGen2.

Note: generative-media models are billed per image / per second / per minute on hosted services — not per token. Running locally or on your own rented GPU is usually far cheaper and keeps your data on your machine.

Key facts

Does	Text → image, Image editing, In context generation, Subject driven generation, Visual understanding
VRAM to run	~17 GB (~17GB on a single RTX 3090/4090 natively; model CPU-offload roughly halves it, sequential offload drops under 3GB at a large speed cost.)
Download	~15 GB
Parameters	7B
License	Apache-2.0 (commercial use OK)
Run with	Diffusers

Get OmniGen2 on Hugging Face →

More multimodal / omni models

Qwen2.5-Omni-7B — Omni understanding + speech generation, ~24 GB VRAM
Janus-Pro-7B — Unified understanding + image gen, ~16 GB VRAM
MiniCPM-o 2.6 — Omni understanding + speech generation, ~9.0 GB VRAM
BAGEL-7B-MoT — Unified understanding + image gen + edit, ~24 GB VRAM
Emu3.5 — Any to any world model (gen + edit), ~48 GB VRAM
Emu3-Gen — Next token any to any generation, ~18 GB VRAM

Browse: all media models · chat / LLM models