# OmniGen2 AI Image Generator and Editor

> Open multimodal generation for text-to-image, instruction editing and combining up to three references.

Canonical URL: https://allimageai.com/models/omnigen-2/

Website: All Image AI

Language: English

OmniGen2 is VectorSpaceLab’s open unified generator. It can create from text, follow an edit instruction or combine people, objects and scenes from as many as three ordered images through one exact endpoint.

## Model details

- Developer: VectorSpaceLab
- Status on All Image AI: Active
- Last verified: 2026-08-25
- Maximum output: 1K
- Modes: Text to Image, Image to Image, Edit Image, Multi-reference
- Reference images: up to 3
- Resolutions: 1K
- Aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3

## Best for

- In-context generation
- Combining people and objects
- Instruction-guided editing
- Open-model workflows

## How to use

1. Upload up to three references.
2. Number every image in the prompt.
3. Balance text and reference guidance.
4. Generate one output before increasing the batch.

## Settings explained

### Reference Guidance

Use 1.3–2.0 for editing and 2.0–3.0 for stronger in-context reference following.

### Text Guidance

Higher values prioritise the written instruction over variation.

## Prompt examples

### Multi-reference

> Put the bird from image 1 on the desk in image 2, use the room lighting from image 3, preserve the bird markings

### Edit

> Keep the person and pose unchanged; make the dress cobalt blue with a matte silk texture

## Limitations

- 1K output
- Slower 50-step default
- Reference-heavy prompts must identify every image by number

## Pricing and credits

Credits for one image at default quality and mode. Resolution, quality, reference inputs, edit mode and output count can change the total. Check the Generate button before submitting.

| Resolution | Credits / image |
| --- | --- |
| 1K | 12 |

Failed generations cost 0 credits. Catalog availability does not guarantee runtime availability; the workspace reports whether the selected model can run.

[Credit plans](https://allimageai.com/pricing.md)

## Test results

Testing in progress. No verified benchmark scores have been published for this model.

## Commercial use and image handling

Commercial use is subject to the OmniGen2 model licence and the selected service terms.

Uploads are processed by the configured inference service. You can delete a generation from your history.

## Frequently asked questions

### Does it need a separate edit endpoint?

No. The same OmniGen2 operation accepts zero to three input images.

### Why does it cost more than draft models?

The verified route bills per output megapixel at a materially higher rate.

## Release history and sources

June 16, 2025 — Open weights released

This date applies to the event shown. Later API access, weights and availability on All Image AI may have different dates.

- [github.com](https://github.com/VectorSpaceLab/OmniGen2)

[Event evidence on the timeline](https://allimageai.com/timeline/#event-omnigen-2)