Midjourney v5 shifts image generation toward realism and finer detail. Compared with v4, the model can interpret a prompt in a way that more closely matches what the user describes, even when the prompt does not specify a visual style. That added flexibility also puts more weight on writing clear, detailed instructions.
More realistic results from the same prompt
Midjourney says v5 was developed for about five months and trained on an “AI supercluster” in the Google Cloud. It uses a significantly modified neural architecture and new aesthetic techniques, with the aim of producing more realistic images and details that are more likely to be correct.
A comparison using the prompt “a tree made of money” illustrates the shift. V4 produces an illustration, while v5 favors a photorealistic installation in a museum. The newer result is also closer to the apparent intent of the prompt, despite the absence of an explicit style instruction.
The difference appears in other examples too. A prompt describing a corporate building entrance with green paint spilling onto the street yields a more illustrative rendering in v4 and a photorealistic one in v5, even without asking for a photograph. Portraits of famous people also look more realistic and contain fewer image errors in the comparison described by the article.
More creative range means more specific prompts
Midjourney founder David Holz calls v5 the “professional mode.” Unlike earlier versions, it is less restricted to particular artistic styles and can produce a wider variety of image results. That freedom can help users explore different interpretations, but it can also make the outcome less predictable if the prompt leaves important choices open.
Holz says users may need longer prompts that spell out details such as lighting and mood. In practice, someone who wants a particular composition or visual treatment may have to describe those choices directly instead of relying on a default style to fill in the gaps.
The change is relevant for both experimentation and repeat work. A short prompt can still produce a coherent image, but the model’s stronger tendency toward photorealism may not match every creative goal. More explicit wording gives the user a way to steer the result, while the broader range of outputs makes it useful to consider which details matter most before generating an image.
The model is still in alpha
The available v5 model is an alpha version. Holz says it will undergo “significant changes” before the final release, and that the finished version will have a more beginner-friendly default style, as previous models did.
That means creators should be cautious about treating the current v5 look as a stable setting for future projects. Work made with the alpha could reflect a style that changes before the final version arrives. The article also notes that hands can still be rendered incorrectly, although extremities are more accurate overall than in v4.
How to try Midjourney v5
Users can select “MJ v5” in Midjourney’s Discord settings or add the parameter “--v 5” to a prompt. The output is directly at double the resolution of v4. Upscaling is not yet available, so the larger output comes from the generation itself rather than a separate upscaling step.
For anyone comparing versions, the clearest approach is to keep the prompt the same and examine how each model interprets it. V5’s more realistic default, increased detail, and broader creative range mark a meaningful change, while the alpha status and occasional rendering errors remain part of the current experience.