Skip to main content
The API provides multiple model version options covering the v6 series, v7 series, and niji series. Each model version has unique advantages in image generation quality, feature characteristics, and parameter support. Supported Models: The API currently supports models v6, v6.1, niji 6, v7, v8, and v8.1, with v7 as the default. Requests without a --v flag run on v7 — add --v 8 or --v 8.1 explicitly to use the newer models.
Midjourney’s own web default moved to v8.1 on June 10, 2026, but the API default remains v7 so existing integrations keep behaving exactly as before. The version is always your choice — pin it explicitly with --v whenever it matters.

​
Version 8.1 🆕

Version 8.1 (alpha launched April 14, 2026) refines V8 with restored image prompting and improved speed, and offers native 2K output via --hd. Add --v 8.1 to your prompt to use it.
Midjourney plans to sunset V8.0 approximately 2 weeks after V8.1 stabilizes. We recommend migrating --v 8 prompts to --v 8.1 ahead of time.

​
Key Improvements over V8.0

  • Native 2K via --hd: default output is 1024×1024 (no separate upscale step needed); add --hd for 2048×2048 at 1.5× cost (120 credits)
  • ~40% faster than V8 HD: V8.1 HD generates at roughly the speed V8 SD did
  • Simple pricing: V8.1 SD is 1× (80 credits); --hd (native 2K) is 1.5× (120 credits) — no other premium tiers
  • Image prompts and image weights restored: V8.1 supports image URLs in the prompt plus --iw for weighting (these were unavailable in early V8.0)
  • Prompt Shortener: Prompts that exceed length limits are automatically shortened by Midjourney before generation
  • Improved Describe: Longer, more detailed prompt descriptions match V8’s prompting style

​
V8.1 Parameters

  • --ar — aspect ratio
  • --s / --stylize — stylization strength (0–1000)
  • --c / --chaos — variability (0–100)
  • --w / --weird — weirdness (0–3000)
  • --raw / --style raw — reduced default styling
  • --exp — experimental aesthetic
  • --sref + --sw — style reference URLs and weight
  • --p — personalization code
  • --iw — image weight (with image URLs in prompt)
  • --hd — native 2K (2048×2048) output, 1.5× cost (120 credits)
  • --no — negative prompt (exclude elements)
  • Moodboards

​
V8.1 Cost Model

  • SD (default): 1x credits (80 points / $0.08 per task)
  • --hd (native 2K): 1.5x credits (120 points / $0.12 per task)
  • --q and --sv 6 are not supported
  • Subscription required: Developer (Pro) plan or above — same as V8
  • Turbo mode (--turbo) is not supported

​
Removed in V8.1 (will be rejected)

(--hd is not rejected — it produces native 2K output at 1.5× cost. See below.) Requests containing rejected flags receive an HTTP 400 with a message naming the offending flag. --no (negative prompt) is supported on v8.1 — it was not removed.

​
Feature Support

​
Version 8

Version 8 (alpha launched March 17, 2026) is a generational leap in AI image generation. Add --v 8 to your prompt to use it.
Midjourney plans to sunset V8.0 approximately 2 weeks after V8.1 stabilizes. New integrations should target --v 8.1.

​
Key Improvements

  • ~5x faster generation: Standard images complete in under 15 seconds
  • Native 2K HD resolution: Use --hd for natively rendered 2K images — no upscale needed
  • Dramatically better text rendering: Wrap text in "quotes" for readable signs, labels, and typography
  • Accurate hands & anatomy: Correct finger counts, natural proportions, consistent full-body poses
  • Superior prompt adherence: Complex multi-element compositions render with much higher fidelity
  • Enhanced coherence: --q 4 mode maintains visual consistency in complex scenes
  • Physically accurate lighting: Reflections, subsurface scattering, and shadow direction are physically grounded

​
V8 Parameters

  • --hd — Native 2K resolution (4x cost)
  • --q 4 — Enhanced coherence mode (4x cost)
  • --oref — Omni Reference, replaces --cref (handles both person likeness and object form)
  • --sv 6 — Use old (more accurate) SREF version (4x cost)
  • All V7 parameters (--chaos, --weird, --exp, --raw, --sref, --stylize, etc.) are supported

​
V8 Cost Model

  • Standard V8 generation: same cost as V7
  • Premium features (--hd, --q 4, --sv 6): 4x cost
  • V8 does not require upscaling — images are natively high resolution
  • Turbo mode (--turbo) is not supported with V8

​
Feature Support

​
Version 7

Version 7 (released April 3, 2025) achieved significant breakthroughs in image generation. This version dramatically improved the processing accuracy of text and image prompts, while achieving a qualitative leap in image quality—particularly in human anatomy, hand details, and object textures. As core innovations, V7 introduces two revolutionary features:
  • Draft Mode: Supports rapid iterative generation
  • Omni Reference (--oref): Provides more powerful image reference capabilities (replaces v6’s Character Reference), bringing new possibilities to creative workflows

​
Feature Support

The quality parameter in V7 works differently. Please refer to the [Quality Parameter Documentation] for details.

​
Version 6.1

Version 6.1 was released on July 30, 2024, and was the default model until v7 launched (June 2025). It generates images with more coherent effects, more precise details and textures, and is approximately 25% faster than version 6.

​
Version 6

Version 6 was released on December 20, 2023, and served as the default model from February 14 to July 30, 2024. Version 6 enhanced prompt accuracy for longer inputs, improved coherence and knowledge, and provided advanced image prompting and mixing capabilities.

​
Niji 6

The Niji model is a special series within the platform, developed in collaboration with Spellbrush, focusing on Eastern and anime aesthetics and illustration styles. Niji 6 was released on June 7, 2024, with improved Japanese text rendering capabilities, efficient processing of short Japanese text, and even the ability to render simple Chinese characters. It provides better image details, particularly in elements like anime eye structures, and resolves some minor image defects that affected a small portion of previously generated images.

​
Model Parameter Comparison

The following table provides a detailed comparison of the parameters supported by each model version and their value ranges:

​
Notes:

  1. ✓ indicates support for this parameter 🚫 indicates no support for this parameter
  2. v8 model: --hd renders natively at 2K resolution (no upscale needed). --hd, --q 4, and --sv 6 each cost 4x credits. Turbo mode is not supported.
  3. v8.1 model: default output is SD (1024×1024) at 1x (80 credits); add --hd for native 2K (2048×2048) at 1.5× (120 credits). The --q, --cref / --cw, and --oref / --ow parameters are removed and will be rejected. --no (negative prompt) remains supported on v8.1.
  4. v7 model exclusive parameters: draft (Draft Mode), oref (Omni Reference), ow (Omni Reference Weight), and exp (Experimental Parameters)
  5. v8 replaces --cref (Character Reference) with --oref (Omni Reference) which handles both person likeness and object form
  6. Video model exclusive parameters: video (Video Mode), motion (Motion Mode), end (Video End Frame), and loop (Video Loop)
  7. v6.1, v7, v8, and v8.1 models support the no parameter (negative prompts); v6 and niji 6 do not