DALL-E and Firefly are usually bought for working tasks rather than art: an illustration for an article, replacing an object on a product photo, editing an image from a description. Open models do this on your own hardware: images and source files stay inside the company, there is no per-image billing, and the generator can be embedded straight into your own product or back office. The cost is a GPU and an engineer to build the pipeline: the model is only weights, the interface and job queue are yours to build. Accuracy on fine detail, text inside the image and exact brand colors is below a mature cloud service, and part of the output will be rejected. In return the whole process belongs to you, including fine-tuning on your own catalog for a consistent house style.
Image generation and editing, including text in images. Earlier versions allow commercial use; the latest 2.1 is non-commercial only.
Open MIT-licensed image models: generation (I1), instruction-based editing (E1) and the unified O1-Image model that does both.
An all-in-one model: generates, edits and moves an object or person from a photo into a new scene without separate plugins.
A hybrid of a 9B language model and a 7B decoder. Strong at text-heavy images: posters, infographics, slides.
NVIDIA's fast image model: 4K images in seconds, runs even on a laptop GPU. The Sprint version generates in 1–2 steps.
HiDream, GLM-Image, OmniGen and SANA are listed as commercially usable (MIT and Apache 2.0), while the recent Qwen-Image-2.1 was released under a research license. Check the license of the exact version you download before going to production.
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
