A new Stable Diffusion model called Openjourney aims to reproduce some of the polished look associated with Midjourney. Its creators say they trained it with more than 60,000 images generated by Midjourney V4, making the project both a free alternative for image makers and a fresh test of how AI-generated work can be reused.
Why build another image model?
Midjourney is known for turning even short prompts into visually appealing results. That ease of use has helped distinguish it from Stable Diffusion and DALL-E 2, whose results can be rougher or require more careful guidance.
Access to Midjourney comes with a cost once the free trial credit is used. Its subscription starts at ten dollars per month. Stable Diffusion, by contrast, is open source, but its basic version may take more sophisticated prompting to get results comparable to Midjourney, and the look can still differ.
Openjourney tries to bridge that gap by building on Stable Diffusion while learning from Midjourney V4 generations. The practical appeal is straightforward: users get another model to experiment with, aimed at producing images in a style associated with a paid service.
What the creators say went into Openjourney
PromptHero, a platform that collects prompts and images made with systems including Midjourney, Stable Diffusion and DALL-E 2, released Openjourney on Hugging Face. The team says the model was trained using more than 60,000 Midjourney V4 images.
The creators have not explained where those training images came from. The source article suggests they may have come from PromptHero’s own platform, but presents that only as a possibility. That leaves an important part of the model’s provenance unclear: the public description identifies the type and volume of material, but not its specific origin or permissions.
PromptHero also opened a separate category for users to share images generated with Openjourney, including the prompts used to make them. The same developers had previously published Dreambooth, which the article reports receives more than 400,000 downloads per month.
Training data brings rights questions
Openjourney’s release adds another layer to an existing dispute over AI training data. Midjourney has faced criticism for training its models on millions of copyrighted photos and artworks without asking their creators. A lawsuit by Getty Images against Stability AI, the company behind Stable Diffusion, is underway.
The question around Openjourney is different in its immediate details: can images generated with one AI service be used to train another model? Midjourney’s terms of service say paying users own all rights to images they create. The article also reports that Midjourney moderators on Discord said users would have to be asked for permission before their images were turned into training material.
That position sits alongside criticism of Midjourney’s own training practices. The company has been accused of using artists’ work without permission to develop and market its model. If AI-generated images are then used to train a separate system, the rights of the artists whose work informed the original model may be implicated again.
Why the source of training images matters
Openjourney highlights how AI models can be built in stages: a model generates images, and those outputs may later become training examples for another model. The process makes it harder to discuss rights by looking only at the final set of generated pictures. Questions also arise about the material that shaped the first model and whether people whose work contributed to it have a say in reuse.
For users, Openjourney offers a route to explore a Midjourney-inspired aesthetic through an open source model. For the broader debate, the unanswered question about where its training images came from is central. A clear account of provenance and permission would help explain what was used and on what basis, while the published description leaves that account unresolved.