Create images from text with Midjourney 8.2, 8.1, 7, or Niji 7. Start with a scene, choose your framing, and tune the visual direction.
Fast generation typically returns four images per task. Page artwork is illustrative; run a prompt to see your model results.
Image generation
One generation typically returns four images. The amount is per task, independent of the number of returned images.
Similar Models
High-fidelity image generation and reference-guided editing with photorealistic detail, typography, and flexible framing.
ByteDance image generation and editing for cinematic visuals, product shots, precise local edits, and reference-guided creative work.
ByteDance image generation and editing for reference consistency, multi-image composition, typography, and polished visual creatives.
Model overview
Develop scenes, illustration, and design directions from a written brief. Choose a Midjourney version in the Playground and use the same parameters through the API. The artwork on this page illustrates creative directions rather than verified model outputs.
Describe a subject, environment, camera angle, materials, and lighting. Use the aspect ratio to frame a wide landscape, portrait, or square composition.
Select Niji 7 for anime and illustrative directions. Specify character details, linework, palette, atmosphere, and background elements in your prompt.
Adjust Stylize for artistic interpretation, Chaos for variation, and Weird for unconventional ideas. Turn on Tile to request artwork that repeats across its edges.
Model characteristics
Choose 8.2, 8.1, 7, niji7. New requests default to 8.2.
Use Raw for more direct prompt control and a negative prompt to describe unwanted elements. A seed helps compare experiments without guaranteeing identical results.
A task typically produces four images. Browse the returned images and download the selected result.
Submit a prompt to receive a task_id, then poll the task endpoint or receive a webhook when it finishes.
FAQ
The version options are 8.2, 8.1, 7, niji7. The default is 8.2; niji7 selects Niji 7.
Start with the subject and action, then describe the setting, lighting, composition, color, and medium. Change one control at a time when comparing results.
Choose 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9. The default is 1:1.
Weird accepts 0–3000 and encourages unusual visual ideas. A fixed seed may not reproduce results when using Weird. Tile requests a repeatable pattern unit in versions 8.2, 8.1, 7. Niji 7 does not support Tile. Both are off by default.
Use the image selector in the result preview or read the complete data.output_urls array from the task API. Generations typically return four images.
