.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
How Midjourneys Discord Based Pricing Rewrote SaaS Monetization Rules
For most of the SaaS era, the product, the account and the commercial model lived inside the same walls. A vendor built its own interface, created user accounts, sold access to those accounts and usually made the seat the unit of expansion. Midjourney broke that sequence. Before August 2024, Discord was its only route for account creation; users prompted a bot inside channels they were already using for conversation and community. Yet Midjourney did not turn Discord itself into the billing system. Subscriptions remained a separate commercial relationship, and today checkout runs through Midjourney's site using Stripe.
The distinction matters because Midjourney's contribution to SaaS pricing was larger than putting a bot in a chat window. It showed that an AI company could separate where customers work from how customers pay, let a third-party network carry much of the early interaction and community experience, and charge for the scarce resource beneath the product rather than the number of people touching it. By 13 August 2026, its four plans still revolve around Fast GPU time, slower unlimited Relax use on higher plans, generation speed, concurrency and privacy rather than conventional feature-seat bundles.
Monetizely's position is that Midjourney rewrote the SaaS playbook by making consumption capacity the primary economic meter while treating the interface as replaceable. That was the right architecture for its first phase. Its next pricing reset should preserve pooled consumption as the primary meter, but move business customers away from personal-account subscriptions towards organisation-level credits, controls and billing.
Discord collapsed acquisition, product use and community into the same loop
Discord gave Midjourney something a conventional SaaS company normally spends years assembling. Discovery could happen in a shared community, product use could happen through the bot, examples from other users appeared beside the workflow, and support could take place in many of the same channels. Midjourney still describes much of its internal communication as taking place on Discord and says product features are suggested and prioritised by its community. As of August 2026, the company describes itself as a lean, self-funded team of about 60 people.
Yet the important pricing lesson is what Midjourney did not outsource. Discord Nitro never entitled a customer to Midjourney, and Midjourney subscriptions do not appear in Discord's subscription settings. Billing belongs to Midjourney. In other words, the company borrowed the interaction layer without surrendering the customer payment relationship.
The progression from Discord-first product to a broader web product shows how the model developed around the meter rather than around a single interface.
| Date | Pricing or product change | What changed economically |
|---|---|---|
| 7 Mar 2024 | Turbo generation was launched at roughly twice the Fast GPU cost for about 3.5 times faster generation. | Speed itself became something users could spend more capacity to buy. |
| 13 Mar 2024 | Midjourney halted free-GPU rewards for image ranking after reporting extensive automated abuse, potentially affecting as much as half of rewarded activity. | A valuable consumption unit attracted gaming as soon as it could be earned. |
| 21 Mar 2024 | Free Fast GPU rewards returned for the top 2,000 daily raters, with new bot detection and abuse controls. | The reward survived, but only after the company strengthened measurement and enforcement. |
| 23 Aug 2024 | Midjourney opened its web platform to everyone; users could create there and connect Discord history. | The product stopped depending on Discord as the only creation surface without changing the core consumption logic. |
| 18 Jun 2025 | V1 video launched at about eight times the cost of an image job, with four five-second videos per job and Relax video testing for Pro and above. | New output types were mapped back to the same capacity system instead of receiving a separate subscription model. |
| 7 Aug 2025 | HD video launched at roughly 3.2 times the cost of normal Midjourney video and was restricted to Pro and Mega. | Higher compute cost became both a higher burn rate and a packaging gate. |
| 13 Aug 2026 | Basic, Standard, Pro and Mega plans remain priced at $10, $30, $60 and $120 monthly, with increasing Fast GPU allocations. Extra Fast time is $4 per hour. | Subscription tiers still pre-purchase different amounts of priority compute rather than additional human seats. |
The sequence shows the deeper innovation. Discord was the distribution and interaction breakthrough, but GPU capacity was the commercial constant. As the interface, models and media types changed, the meter survived.
That distinction has become more visible since the web product matured. Every current subscription can be used on both the website and Discord, while settings such as generation speed synchronise between them. A company copying Midjourney today therefore should not copy "Discord pricing". It should copy the separation of interface, identity, community and monetisation.
GPU time made costly AI use scalable without charging for every human
A seat would have been an unusually weak primary metric for Midjourney. Two designers could consume vastly different amounts of inference, while one intensive user creating video could place more load on the service than a small team making occasional images. Current Midjourney documentation makes the cost relationship explicit: subscribers are buying time on GPUs, and the amount consumed varies with the job.
As of 13 August 2026, a normal SD image prompt is listed at about 0.8 GPU minutes, an HD image prompt at roughly 1.3 minutes, four SD videos at about eight minutes and four HD videos at around 26 minutes. Turbo mode can run supported images up to four times faster while using twice the Fast time. The customer therefore pays more when the product does more expensive work, even though Midjourney hides the underlying hardware bill inside a simpler unit.
The current plan architecture makes that relationship visible.
| Plan, 13 Aug 2026 | Monthly price | Fast GPU time | Relax entitlement | Commercial/package signal |
|---|---|---|---|---|
| Basic | $10 | 3.3 hours | None | Low-cost entry with commercial terms |
| Standard | $30 | 15 hours | Unlimited images | Heavy image exploration becomes affordable without unlimited priority compute |
| Pro | $60 | 30 hours | Unlimited images and SD video | Adds Stealth and higher concurrency; required at minimum for companies above $1m annual gross revenue |
| Mega | $120 | 60 hours | Unlimited images and SD video | Highest included capacity and video concurrency |
| Extra Fast time | $4/hour | Added as purchased | N/A | Overage lets high-intensity users expand without changing the basic meter |
Source: Midjourney's official plan documentation, accessed 13 August 2026. Annual billing receives a 20% discount.
The table reveals why Midjourney's model is more sophisticated than simple usage pricing. Fast time is scarce and metered; Relax time is deliberately abundant. A Standard subscriber can generate unlimited images in Relax mode, but must spend a limited pool when speed matters.
That split lets Midjourney anchor price partly to cost and partly to value. Compute cost explains why HD video burns more. Customer urgency explains why Fast and Turbo matter. Privacy supplies another value lever: Stealth is reserved for Pro and Mega even though hiding a creation is not plausibly 6 or 12 times as costly to compute as the Basic plan.
Monetizely's view is that this combination is the enduring lesson. Pure cost-plus pricing would sell GPU minutes like a cloud provider. Pure value pricing would attempt to charge differently for an advertising concept, a game asset and a personal illustration even when Midjourney cannot reliably observe the value of any of them. Capacity sits between those extremes.
The weakness appears as the product adds media types. "GPU hour" is technically consistent but less intuitive when an SD image batch, an HD image, an upscale and an HD video all consume it at different rates. What began as a simple proxy for generations is becoming a conversion table that customers must learn. Midjourney should retain consumption, but eventually present the customer-facing meter as a common credit whose burn rate is shown before each job.
The five-step diagnosis gives the meter high marks and the business layer a failing grade
Monetizely's 5-Step Pricing Framework treats pricing as a sequence rather than a hunt for the perfect number. Goals and Segmentation establishes which customers the company wants and what pricing must accomplish. Packaging decides which capabilities, limits and service levels belong together. Pricing Metric chooses the unit that makes the bill expand. Rate Setting puts a price on that unit after testing value, willingness to pay and cost. Operationalization makes the design work in entitlements, metering, quoting, billing and customer reporting. The order matters because a clever meter cannot compensate for a package aimed at the wrong buyer, and no model survives if finance cannot explain the invoice. The fuller treatment appears in Monetizing Agentic AI.
For Midjourney, the most revealing steps are packaging, metric and operationalisation. The grades are intentionally uneven.
| Framework step | Grade | One-line rationale |
|---|---|---|
| Packaging | B | The creator ladder cleanly separates capacity, speed, privacy and concurrency, but it does not become a real organisation package when a company needs central control. |
| Pricing Metric | A- | GPU time tracks variable delivery cost and heavy use far better than seats, while Relax mode stops usage pricing from punishing exploration; growing workload differences make the unit harder to understand. |
| Operationalization | C+ | Individual users can see usage, buy more time and manage subscriptions, but business customers still lack core account administration, SSO and procurement support. |
The strongest part is metering. In Discord, /info shows a subscriber's plan, renewal date, remaining Fast time, lifetime activity and Fast, Turbo and Relax usage. Usage visibility therefore sat inside the workflow even before the website became the broader creation environment.
The biggest weakness is equally concrete. As of 13 August 2026, Midjourney says organisations buying fewer than 50 annual Pro or Mega subscriptions cannot obtain a single invoice for multiple accounts. Even at 50 or more, the company says it does not provide seat or admin management, SSO, terms negotiation, or participation in vendor and security processes. Each account still requires its own billing email, and Midjourney advises organisations to keep a spreadsheet of users, emails and account IDs.
For a creator product, that simplicity can be an advantage. For a marketing department with 300 designers, agencies and contractors, it becomes a procurement problem. A finance team should not need an external spreadsheet to answer who can spend company money on an AI service.
Midjourney therefore gets the hard economic choice mostly right and the mundane enterprise plumbing wrong. SaaS companies often make the reverse mistake.
Midjourney sits too close to the human creator for outcome pricing to make sense
AI does not automatically imply outcome pricing. The Agentic Monetization Spectrum, or AMS, separates products along three dimensions. Zero-Human Ability asks how much work remains with a person: Small means the human still does at least half, Medium means the person delegates and reviews, and Large means the system performs almost all of the work. Operational Domain moves from one task, to an end-to-end function, to work spanning several functions. Output/Cost Ratio asks whether output value rises roughly with compute cost, begins to pull materially ahead of it, or overwhelms it. As autonomy, scope and the value-to-cost gap grow, price can move away from human access and towards completed work or outcomes.
Midjourney lands well short of the outcome end of that spectrum.
| AMS dimension | Midjourney assessment | Evidence behind the score | Pricing implication |
|---|---|---|---|
| Zero-Human Ability | Medium | The model generates the asset, but a person still supplies intent, selects results, reruns prompts, edits and decides what is usable. | Seats are too human-centric, but the final business outcome is still too dependent on the user. |
| Operational Domain | Small | Midjourney creates and edits visual media; it does not independently run an end-to-end marketing, design or production function. | Charge for creation capacity rather than a department-level outcome. |
| Output/Cost Ratio | Inflecting | A commercially useful image can be worth far more than its compute cost, yet many generations are discarded and the value of the chosen asset varies widely. | Value-based packaging can sit above a consumption meter, but pure cost pricing leaves money behind and outcome fees reach too far. |
The AMS therefore reinforces the existing architecture rather than pointing away from it. Consumption is the right centre of gravity. Per-seat pricing would charge the same whether a designer generated five images or 5,000; per-outcome pricing would force Midjourney to decide whether a campaign asset "worked", even though distribution, copy, media spend and the customer's own decisions may matter more than the generated image.
Video does not overturn that logic. When Midjourney released V1 video on 18 June 2025, its official update acknowledged that production cost and price were hard to predict and said the company would adjust access to ensure it was "operating a sustainable business." That is precisely the environment in which consumption should remain close enough to cost to stop successful power users from becoming unprofitable.
An AI agent that independently produces and runs a complete campaign would occupy a different AMS position. Midjourney, as sold in August 2026, remains a powerful creative producer directed by a human rather than an autonomous owner of the business result.
The market copied consumption while exposing the danger of opaque credits
Midjourney was early to a pattern that is now visible across creative AI and AI-enabled SaaS: a recurring subscription buys a quantity of expensive AI work, while heavier customers either buy more consumption or move into a higher tier. The implementation differs, but the economic direction is remarkably consistent.
The comparison matters because it shows both what Midjourney anticipated and what later companies have improved.
| Company and primary source date | Current or documented meter | What the model says about Midjourney's choice |
|---|---|---|
| Midjourney, 13 Aug 2026 | Subscription plus Fast GPU hours; unlimited slower image generation on Standard and above; $4 per additional Fast hour. | Strong cost linkage, but the customer sees an infrastructure-flavoured unit. |
| OpenAI, 23 Apr 2025 | gpt-image-1 API billed by text-input, image-input and image-output tokens; OpenAI gave approximate per-image costs of $0.02, $0.07 and $0.19 at three quality levels at launch. |
Developer products can expose granular infrastructure usage because software systems, rather than end users, reconcile the bill. |
| Runway, 13 Aug 2026 | Standard plan at $15 monthly, or $12 monthly when billed annually, includes 625 credits; those credits translate into different quantities of image and video generation. | Credits provide a common language across workloads even when underlying compute differs. |
| Adobe Firefly, 13 Aug 2026 | Paid Firefly tiers bundle monthly generative credits; business plans add admin tools and support, with higher allowances for heavier creative work. | Adobe layers enterprise administration on top of consumption instead of asking each user to manage an isolated account. |
| Figma, 13 Aug 2026 | Seats include monthly AI credits, while teams can buy shared credit pools or pay as they go; Professional Full seats list 3,000 monthly credits, Organisation 3,500 and Enterprise 4,250. | A mature SaaS vendor can keep seats for collaboration while moving marginal AI consumption onto a second meter. |
| Unity, 12 Sep 2024 | Unity cancelled its planned Runtime Fee and returned gaming customers to seat subscriptions after customer consultation. | A theoretically measurable usage event still fails when customers distrust the event or cannot comfortably forecast the bill. |
The market evidence supports Midjourney's consumption thesis, but it also shows where Midjourney now trails. Runway calls the common unit a credit. Figma has moved additional AI consumption into shared team pools. Adobe surrounds generative usage with enterprise administration. Midjourney still asks business customers to manage many essentially personal subscriptions.
Practitioners have been unusually candid about the tension.
In a Midjourney update attributed to Max Ingham on 13 March 2024, the company halted free Fast-time rewards because the reward appeared to be "incentivizing bad behaviour." The episode is an important pricing lesson: once a scarce unit has monetary value, every free path into that unit becomes part of the pricing system and needs fraud controls.
Figma encountered the opposite problem after moving from generous AI access towards enforced consumption. Its Form 10-Q for the quarter ended 31 March 2026 says that, after AI credit limits began to be enforced in March, Figma observed higher support volume, public customer dissatisfaction and reduced usage by some customers. The same filing says its seat model may become less aligned with value as AI automates work previously requiring multiple users.
Figma CFO Praveer Melwani described the company's monetisation goal on its 14 May 2026 earnings call as to "support adoption rather than constrain it." By the 5 August 2026 call, Figma was adding user-level controls and adapting the model as customers learned to purchase, allocate and govern credits. CEO Dylan Field captured the state of the market neatly: "we are learning a lot and so is the market generally."
Adobe offers the counterexample on packaging. When it commercially launched Firefly in September 2023, it embedded Generative Credits into subscriptions rather than treating AI as a separate infrastructure product. David Wadhwani, then president of Adobe's Digital Media business, argued that Firefly and Adobe's existing tools could give creators "unparalleled opportunities to work with generative AI." By August 2026, Adobe's Firefly business plans combine credits with admin tools and support.
These examples point to the rule Midjourney should keep: meter expensive work, but make the meter easier to buy than the infrastructure beneath it.
Pooled business credits should be Midjourney's next pricing reset
Midjourney should not retreat to seats simply because it now has a polished web application. The cost problem that justified GPU-time pricing remains. Nor should it jump to outcome pricing because its images can create commercial value. The human creator still determines what gets made, selected, edited and deployed.
The next reset should be more specific: replace GPU hours with a customer-facing credit unit for business plans, pool those credits at the organisation level, and keep consumption as the primary expansion metric. Midjourney can continue converting credits internally into GPU demand at different rates for images, SD video, HD video, upscaling and future workloads.
Such a reset would also solve the enterprise gap without corrupting the core economics. A company could buy £50,000 or $100,000 of annual capacity, distribute access across employees, and let usage move between people and projects. Administrators could see who consumed what without making each person a revenue unit.
The product already proves that identity and creation surface can change while the meter persists. Discord-only account creation gave way to web and Google sign-in after August 2024; current subscribers can work on either interface. The commercial architecture should now make the same transition from individual creators to organised teams.
For SaaS and AI operators studying Midjourney, five decisions follow:
Separate your primary meter from your interface. A customer who moves from web to Slack, Teams, an API or an agent should not force the company to redesign its entire revenue model.
Choose one consumption unit that can survive new workloads. Keep tokens, GPU seconds and model calls inside engineering where possible; expose a stable credit or completed-work unit that customers can forecast across product generations.
Decide explicitly whether the business is monetising scarcity or value. Midjourney should continue monetising expensive generation through consumption while capturing extra value through package gates such as privacy, concurrency and service levels.
Build the organisation account before chasing the largest enterprises. Shared budgets, SSO, role controls, project allocation and one invoice are not sales extras. They determine whether a consumption model can move from individuals to procurement-led companies.
Protect exploration from the meter. Midjourney's Relax mode is strategically important because every experiment does not feel like a taxi meter running. AI companies should preserve a low-cost or slower path for discovery while charging more when speed, scale or premium models create real incremental cost.
Midjourney's most important monetisation idea was never that SaaS should move into Discord. Discord was a remarkably effective starting point, but Midjourney itself has already moved beyond it. The larger contribution was to show that the user interface can be temporary while the economic unit remains durable.
Traditional SaaS sold access to software and let marginal use cost almost nothing. Generative AI reversed that assumption. Midjourney responded early by selling a recurring claim on scarce compute, then using speed, privacy and abundance to turn that capacity into packages customers could understand.
The company now needs to complete its own lesson. Keep consumption primary, abstract GPU time into credits, pool those credits for organisations, and make Discord one excellent place to work rather than the shape of the account. That would not abandon the model that rewrote the rules. It would make the model ready for the market that followed.
Assumptions
Pricing and product features described as current were checked against official sources on 13 August 2026. Midjourney describes itself publicly as self-funded and does not publish the 10-K and earnings-call materials available for listed companies; SEC material is therefore used only where another public company makes a relevant filing. A verifiable Wayback capture of Midjourney's historical pricing page was not available through the research environment, so no archived price was inferred or fabricated; dated historical statements rely on Midjourney's own update record. AMS grades and framework grades are Monetizely's analytical assessments, not Midjourney disclosures.
Footnotes
Ajit Ghuman and Akhil Gupta, Monetizing Agentic AI, 2026. https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
Midjourney, "Logging In & Connecting Accounts," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/33390994570509-Logging-In-Connecting-Accounts
Midjourney, "How to Subscribe" and "Discord Nitro Subscription," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/31974654274573-How-to-Subscribe https://docs.midjourney.com/hc/en-us/articles/27870371412749-Discord-Nitro-Subscription
Midjourney, "Comparing Midjourney Plans," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/27870484040333-Comparing-Midjourney-Plans
Midjourney, "GPU Speed (Fast, Relax, Turbo)," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/32016412137741-GPU-Speed-Fast-Relax-Turbo
Midjourney, "Careers" and "About," accessed 13 August 2026. https://www.midjourney.com/careers https://alpha.midjourney.com/
Midjourney, "Launching V6 Turbo," 7 March 2024. https://updates.midjourney.com/launching-v6-turbo/
Midjourney, "Halting free-gpu rewards," 13 March 2024. https://updates.midjourney.com/halting-free-gpu-rewards/
Midjourney, "New Style Refs and the return of free GPU hours," 21 March 2024. https://updates.midjourney.com/new-style-refs-and/
Midjourney, "The web is open to all!", 23 August 2024. https://updates.midjourney.com/the-web-is-open-to-all/
Midjourney, "Introducing Our V1 Video Model," 18 June 2025. https://updates.midjourney.com/introducing-our-v1-video-model/
Midjourney, "HD mode for video," 7 August 2025. https://updates.midjourney.com/hd-mode-for-video/
Midjourney, "Using Midjourney in Discord," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/31541509949069-Using-Midjourney-in-Discord
Midjourney, "Stealth Mode," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/32019750070669-Stealth-Mode
Midjourney, "Info Command," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/32084927086861-Info-Command
Midjourney, "Group Plans & Corporate Billing," accessed 13 August 2026. https://docs.midjourney.com/hc/en-us/articles/27870607078285-Group-Plans-Corporate-Billing
OpenAI, "Introducing our latest image generation model in the API," 23 April 2025. https://openai.com/index/image-generation-api/
Runway, "Pricing" and "How do credits work?", accessed 13 August 2026. https://runwayml.com/pricing https://help.runwayml.com/hc/en-us/articles/15124877443219-How-do-credits-work
Adobe, "Adobe Firefly" plans, accessed 13 August 2026. https://www.adobe.com/products/firefly/plans.html
Figma, "Plans & Pricing," accessed 13 August 2026. https://www.figma.com/pricing/
Unity Technologies, Matt Bromberg, "Unity is Canceling the Runtime Fee," 12 September 2024. https://unity.com/blog/unity-is-canceling-the-runtime-fee
Figma, Inc., Form 10-Q for the quarter ended 31 March 2026. https://www.sec.gov/Archives/edgar/data/1579878/000162828026035209/fig-20260331.htm
Figma, Inc., Q1 2026 Earnings Call transcript, 14 May 2026. https://investor.figma.com/news-events/events-and-presentations/event-details/2026/Figma-Q1-2026-Earnings-Call/default.aspx
Figma, Inc., Q2 2026 Earnings Call transcript, 5 August 2026. https://investor.figma.com/news-events/events-and-presentations/event-details/2026/Figma-Q2-2026-Earnings-Call/default.aspx
Adobe, "Adobe Unleashes New Era of Creativity for All with the Commercial Release of Generative AI," September 2023. https://news.adobe.com/news/news-details/2023/adobe-unleashes-new-era-of-creativity-for-all-with-the-commercial-release-of-generative-ai