Midjourney capabilities

Character-consistent image generation: point Midjourney at a reference image of a character with the --cref parameter and it tries to reproduce that same character (face, and optionally hair and clothing) across new prompts. The companion --cw (character weight) parameter is a 0-100 dial controlling how strongly the reference is applied: --cw 100 (the default) pulls in face, hair, and clothes, while --cw 0 focuses mainly on the face so you can change the outfit or hairstyle through the text prompt. You still write a normal text prompt; the reference only constrains the character's appearance, not the scene.

Describe: reverse-engineers a text prompt from an uploaded image, returning four example prompts that capture the style, composition, and subject. Updated for V8.1 to produce longer, more detailed prompts. Blend: combines 2-5 uploaded images into a single new image by merging their concepts and visual styles.

A unified web-based editor for non-destructive post-generation image editing. Capabilities include inpainting (Vary Region / Erase+Restore: selectively regenerate a masked area), outpainting (Zoom Out: expand the canvas beyond original edges), directional canvas extension (Pan: add content left/right/up/down), and region-specific repainting. Supports undo, redo, and reset.

Post-generation upscaling that doubles image resolution from the default ~1024px to ~2048px. Two modes: Subtle Upscale (preserves original look, minimal changes) and Creative Upscale (doubles size while adding AI-generated detail improvements, can correct artifacts like awkward hands or odd facial expressions).

Animates a still image into a short video clip using Midjourney's V1 video model (released June 2025). Each job produces four 5-second clips at 480p/24fps. Videos can be extended up to 4 times (~4 seconds per extension), reaching a maximum of ~21 seconds total. Accepts both Midjourney-generated images and user-uploaded external images as starting frames.

Midjourney offers four paid subscription tiers with no permanent free plan as of 2026. Plans are differentiated by fast GPU hours per month, Relax mode availability, Stealth mode (private outputs), and concurrent job slots. Annual billing saves 20% (effective monthly rates: $8/$24/$48/$96).

A rich set of end-of-prompt parameters that adjust composition, aesthetics, randomness, and output format. Core parameters include --ar (aspect ratio), --stylize (aesthetic processing intensity, 0-1000), --chaos (output variety, 0-100), --weird (surreal deviation, 0-3000), --no (element exclusion), --seed, --tile (repeating patterns), and --raw/--style raw for reduced stylization.

Three overlapping tools for persistent aesthetic control. Style Reference (--sref <URL or code>): apply the lighting, color grading, and texture of a reference image without copying its content. Personalization (--p): after rating ~200 images, Midjourney builds an aesthetic profile that subtly customizes all outputs to match individual taste. Moodboards: curated image collections that anchor style across a project; highly varied moodboards yield diverse outputs, narrow moodboards yield focused outputs.

Converts a natural-language text prompt into four image variants using Midjourney's diffusion-based models. V8.1 became the default model on June 11, 2026 (replacing V7); V8.0 alpha is being deprecated. V8.1 delivers images in approximately 4 seconds (SD) or 12 seconds (HD), with HD output at twice the linear size and 4x the resolution of V7 images. V7 remains available for users who want the Omni Reference (--oref) feature, which is not supported in V8.1.

Midjourney is accessible via two distinct surfaces: the primary midjourney.com web app (launched 2024, now the recommended interface) and the Discord bot. Both access identical AI models and produce identical image quality; the difference is in workflow, tooling, and community features.