Companies with gross annual revenue over $1,000,000 USD must purchase a Pro or Mega plan to use Midjourney for commercial work. The Pro plan costs $60 per month, whereas the Mega plan costs $120 per month. Both tiers include Stealth Mode to keep images and videos private from the public gallery. The Pro plan provides 30 hours of Fast GPU time per month, but the Mega plan provides 60 hours. Standard users get 15 hours of Fast GPU time and unlimited Relax mode. The Basic plan is $10 per month. Users save 20% on any plan by paying the full year fee upfront with an annual subscription. For the Basic plan, the annual cost is $8 per month. The Standard plan annual cost is $24 per month. The Pro plan annual cost is $48 per month and the Mega plan is $96 per month. If a user exhausts Fast GPU time, they can buy extra time for $4 per hour. Users can access Midjourney through the web interface at midjourney.com or through Discord. V7 text rendering accuracy improved by 15% over V6, but it still maintains only 30% accuracy for complex sentences. The Basic plan is for casual users, while the Mega plan is for studios. Users can rate images to earn free GPU time. The web app allows users to use the Editor for inpainting, outpainting, and retexturing. Fast mode processes tasks immediately with high priority, while Relax mode queues requests based on system availability.
Parameter engineering for visual consistency
Professional users manage consistency through specific parameter commands. They use the –cref tag to point to a URL of a character. The –cw parameter adjusts the weight from 0 to 100. Setting the weight to 100 copies the face, clothes, and hair. Dropping the weight to 20 keeps the face but changes the outfit. To maintain a specific brand identity, a marketing professional adds the –sref command followed by a URL of a brand mood board to force Midjourney to mimic specific color palettes and aesthetic styles for every single generation.
| Plan | Monthly Price | Fast GPU Time | Stealth Mode |
|---|---|---|---|
| Basic | $10 | 3.3 hours | No |
| Standard | $30 | 15 hours | No |
| Pro | $60 | 30 hours | Yes |
| Mega | $120 | 60 hours | Yes |
Users can adjust the aspect ratio using the –ar parameter. For example, –ar 9:16 is used for TikTok covers and –ar 4:5 works for Instagram. The –stylize parameter, or –s, controls artistic liberty from 0 to 1000, with a default value of 100. The –chaos parameter, or –c, ranges from 0 to 100 and changes the variation between the four images in a set. The –weird parameter, or –w, ranges from 0 to 3000. The –style raw parameter removes default beautification. The –sw parameter can go as high as 1000 to increase style intensity. Professionals follow a structure of subject, context, lighting, style, and parameters. You should check the V7 parameters if you want to avoid wasting GPU hours on bad compositions. Users can use image prompts to influence content, composition, style, and colors.
Programmatic access and content control
Developers use CometAPI to access Midjourney models via a pay-as-you-go system. CometAPI provides access to Midjourney V7 and V8. TTAPI provides three modes: fast, relax, and turbo. Fast mode delivers images in 90 seconds. Relax mode takes under 10 minutes. Turbo mode acts as a priority option. The cost for a Fast generation is $0.08 per image, while Turbo costs $0.159. Users can use the /midjourney/image-edits/submit endpoint via TTAPI to perform image edits.
Moderation remains necessary for brand safety. Unmoderated content can expose a brand to offensive material. Companies should use AI and human teams to filter user-generated content. Hive uses a multimodal LLM to evaluate images and text together and includes CSAM detection via Thorn integration. Sightengine detects deepfakes and AI-generated content. Checkstep focuses on governance and auditability for regulatory reporting. Besedo handles marketplace fraud and scams. Cleanspeak includes review queues. Alice provides threat intelligence for coordinated abuse and security for generative AI models. WebPurify uses a hybrid AI and human model for user-uploaded photos. Stream provides real-time, multi-format detection with configurable policy logic.
Will the V7 model eventually reach 100% accuracy for complex typographic requirements?
