Midjourney vs Runway

TL;DR

Midjourney is the still-image aesthetic engine. Runway is the generative video and world-model platform. As of late July 2026, they sit in different lanes more than ever, even as both expand into adjacent categories.

Two product moves bookend this comparison. On July 24, 2026, Midjourney shipped V8.2—a “more creative, bold, sophisticated, edgy and fresh” image model with smarter personalization (midjourney.com/updates/version-8-2). One day earlier, on July 23, 2026, founder David Holz announced Midjourney’s first acquisition: the astrology app Co-Star, with a team of about two dozen joining from Co-Star (updates.midjourney.com/midjourneys-first-acquisition; TechCrunch, July 24, 2026).

Across the same window, Runway is operating at a different scale and a different media type. Its homepage advertises “60M+ creatives around the world” using Runway Creative (runwayml.com). Its flagship video model Gen-4.5, announced December 1, 2025, holds the top spot on the Artificial Analysis Text-to-Video benchmark with 1,247 Elo points as of late November 2025—above Google Veo 3 (1,226), Kling 2.5 (1,225) and OpenAI Sora 2 Pro (1,206) (runwayml.com/research/introducing-runway-gen-4.5).

“Gen-4.5 currently holds the top position in the Artificial Analysis Text to Video benchmark, surpassing all other models.” — Runway, December 1, 2025

How the two products actually differ

Midjourney is a generative image model with a web app, a Discord bot, a 5-second video mode, and a growing stack of side projects (medical imaging, an acquired consumer app). It’s run by a self-funded research lab of about 60 people (midjourney.com) and is best understood as a creative direction tool for stills.

Runway is a generative video and world-model platform. It runs text-to-video, image-to-video, video-to-video, motion capture (Act-Two), an in-app video editor (Aleph 2.0), and an autonomous creative agent (Runway Agent 2.0, June 25, 2026). On the research side, it ships General World Models—autoregressive, real-time simulations of environments—through GWM-1 (December 11, 2025) and three commercial variants: GWM Worlds, GWM Avatars (Runway Characters), and GWM Robotics (runwayml.com/research/introducing-runway-gwm-1).

The two products overlap in “AI creative production,” but they do not solve the same primary problem. If your next deadline is a still, you start with Midjourney. If your next deadline moves, you start with Runway.

Midjourney in July 2026: V8.2, V8.1, and the wider lab

Midjourney’s V8.2 release, dated July 24, 2026, is positioned as an aesthetic and personalization upgrade—not a paradigm shift (midjourney.com/updates/version-8-2). The official post lists four changes:

• Images should be more creative, bold, sophisticated, edgy and fresh
• Random situations where you’d get a low quality image should be dramatically reduced
• Personalization will “understand” your personal taste much better, especially if you have a lot of ratings under your profile
• Creating personalization profiles for V8.2 now has a much larger and improved pool of images to choose from

V8.2 follows V8.1, which became the default model on June 11, 2026. V8.1 was already “smarter, more coherent, better adheres to detailed prompts, and renders text better than ever” and ran at 4 seconds in standard mode and 12 seconds in HD (updates.midjourney.com/v8-1-is-now-the-default-model). The April 30, 2026 V8.1 update also brought Discord parity and improved sharpness for style references and moodboards (updates.midjourney.com/v8-1-updates).

A few other notes from Midjourney’s own product log in June and July 2026:

• Draft mode rolled out for V8.1 on June 16, 2026. Each draft job creates 24 images at lower resolution, using “half as many fast hours as V8.1 SD jobs” (updates.midjourney.com/draft-mode-for-v8-1-and-new-feature-previews).
• Random styles in draft mode shipped June 25, 2026 via the --sref random flag (updates.midjourney.com/random-styles-in-draft-mode).
• A hardware announcement livestream is scheduled for June 17, 2026 at 6pm Pacific (updates.midjourney.com/hardware-announcement-livestream). The official site frames the device as “ambitious, (physically!) big, and unexpected.”
• The first acquisition was announced July 23, 2026: Co-Star, the social astrology app with about 4.3M monthly active users, per TechCrunch (updates.midjourney.com/midjourneys-first-acquisition; TechCrunch, July 24, 2026).

Midjourney’s terms of service, version effective May 27, 2026, confirm that you own the assets you create—with two important caveats. First, “if you are a company or any employee of a company with more than $1,000,000 USD a year in revenue, you must be subscribed to a ‘Pro’ or ‘Mega’ plan to own Your Assets.” Second, by default your content is publicly viewable and remixable unless you pay for Stealth mode (docs.midjourney.com/hc/en-us/articles/32083055291277-Terms-of-Service). That $1M revenue gate is a real friction point for agencies, brand teams, and any startup crossing the eight-figure threshold.

Beyond the core model, Midjourney is investing in two side bets. Midjourney Medical is building “Ultrasonic CT,” a full-body scanner using sound waves and water, with a goal of 50,000 scanners worldwide by 2031 and the first San Francisco spa opening in 2027 (midjourney.com/medical). The acquisition of Co-Star, with its consumer-app DNA, also signals a more productized future for the lab.

Runway in July 2026: Gen-4.5, GWM-1, Lionsgate, and 60M+ users

Runway’s 2026 story is about a model family that is no longer just “AI video.” Three layers matter.

1. The video model: Gen-4.5 and the Text-to-Video leaderboard

Runway Gen-4.5 launched on December 1, 2025. The announcement post is explicit: 1,247 Elo on the Artificial Analysis Text-to-Video leaderboard, ahead of Veo 3 (1,226), Kling 2.5 (1,225) and Sora 2 Pro (1,206). Elo is “determined by blind human preference voting across thousands of pairwise comparisons, with no knowledge of which model produced each video” (runwayml.com/research/introducing-runway-gen-4.5).

A few practical details from the same post:

• Gen-4.5 was developed end-to-end on NVIDIA GPUs and runs inference on NVIDIA Hopper and Blackwell series hardware
• Jensen Huang, NVIDIA’s CEO, is quoted endorsing the partnership
• Pricing is “comparable across all subscription plans” with Gen-4.5 included on every paid Runway plan
• Existing Gen-4 controls—Image to Video, Keyframes, Video to Video—carry over to Gen-4.5

Gen-4.5 is also the foundation for GWM-1, Runway’s general world model announced on the same day, December 11, 2025 (runwayml.com/research/introducing-runway-gwm-1).

2. The world model layer: GWM-1, Characters, Robotics, Worlds

GWM-1 is an autoregressive model that “generates frame by frame, runs in real time, and can be controlled interactively with actions—camera pose, robot commands, audio.” It ships in three post-trained variants:

• GWM Worlds — explorable, real-time environments for gaming, training agents, VR
• GWM Avatars — Runway Characters, audio-driven interactive avatars with realistic facial expressions and lip-sync
• GWM Robotics — a video-generation backbone for robot policy models, with synthetic data augmentation and offline policy evaluation

Runway Characters, the consumer-facing version of GWM Avatars, launched on March 9, 2026, and is already used by partners including BBC and Silverside (runwayml.com/news/introducing-runway-characters).

3. The product surface: Agent 2.0, Aleph 2.0, Studio, third-party models

The Runway changelog through July 2026 reads like a product org sprinting across modalities (runwayml.com/changelog):

• Jun 25, 2026 — Agent 2.0: “Runway Agent creates entire campaigns, analyzes data to improve your creative and scales your marketing across platforms, formats and markets.”
• Jul 1, 2026 — Nano Banana 2 Lite for fast image generation
• Jul 2, 2026 — Agent Skills, a command-based way to run ad campaigns, commercials, and localization inside Runway
• May 21, 2026 — Aleph 2.0, an upgraded video editing model
• May 13, 2026 — Runway Agent debuts as an AI creative partner
• Apr 7, 2026 — Seedance 2.0 arrives for Unlimited and Enterprise
• Mar 9, 2026 — Runway Characters goes live
• Feb 20, 2026 — Third-party model hub: Kling 3.0, Kling 2.6 Pro, WAN2.2 Animate, GPT-Image-1.5, Sora 2 Pro, and more
• Jan 21, 2026 — Gen-4.5 Image to Video

Runway’s pricing page currently advertises “Trusted by 60M+ creators and leading enterprises” (runwayml.com/pricing). The same page lists four plans on the Creative tier: Free (one-time 125 credits), Standard $12/mo annual ($15 monthly), Pro $28/mo annual ($35 monthly), and Max $76/mo annual ($95 monthly). Enterprise is custom.

The Lionsgate relationship is the clearest production proof point

Both companies have marquee customers, but Runway’s Lionsgate relationship is unusually concrete. The original partnership was announced September 18, 2024 as a “first-of-its-kind” deal to train a custom AI model on Lionsgate’s catalog (runwayml.com/news/runway-partners-with-lionsgate).

It expanded on June 11, 2026. Lionsgate took an equity stake in Runway, and the two companies agreed to a joint development program for new IP, beginning with a short-form episodic series (runwayml.com/news/runway-and-lionsgate-expand-partnership). From the post:

“Lionsgate is the first Hollywood studio to partner with an applied research AI company, the first to hire a Chief AI Officer and the first to build out an infrastructure to support its AI strategy.”

You can see Lionsgate DNA inside Runway’s own customer page. House of David, an Amazon Prime series produced by Kingdom Story Company and Wonder Project, used Runway for Episode Six’s Goliath origin sequence. The customer story cites 432 minutes of airtime across 8 episodes, 22 million viewers in the first 17 days, 5 months of post-production time saved, and a #1 finish on Amazon Prime (runwayml.com/customers/how-house-of-david-used-runway-to-become-amazons-latest-hit-series). The director, Jon Erwin, says the AI workflow let his team get through a “couple weeks” of complex VFX that would have taken “four or five months in a traditional process.”

For Midjourney, there is no equivalent Hollywood case study in the public record. Midjourney’s most-cited production usage remains brand, editorial, and concept art—not feature-film VFX pipelines.

Pricing, credits, and the ownership fine print

Both companies charge a subscription, but the billing logic is different.

Runway pricing (runwayml.com/pricing, July 2026)

• Free — one-time 125 credits, 3 editor projects, 5GB storage
• Standard — $12/mo annual, 625 credits/mo, all video and image models, 4K upscaling, no watermarks
• Pro — $28/mo annual, 2,250 credits/mo, 500GB storage, audio tools, “best models”
• Max — $76/mo annual, 9,500 credits/mo, 1-month credit rollover, first access to newest models
• Enterprise — custom credits, SSO, workspace analytics, configurable teamspaces, priority support

Gen-4.5 costs 60 credits per 5 seconds. Gen-4 Image at 1080p is 8 credits per image. Aleph 2.0 is 140 credits per 5 seconds (runwayml.com/pricing).

Midjourney pricing

Midjourney’s full plan ladder isn’t reproduced in the public changelog, but the May 27, 2026 terms confirm three things you should know before signing a contract (docs.midjourney.com/hc/en-us/articles/32083055291277-Terms-of-Service):

1. Plans include Basic, Standard, Pro, and Mega tiers (referenced in the Stealth Mode section)
2. Pro and Mega unlock Stealth (private outputs) and are required for asset ownership if your company earns more than $1,000,000 in annual revenue
3. You own the assets you generate, and ownership persists if you downgrade or cancel

In practice, expect Midjourney to be the cheaper choice for high-volume still-image iteration and Runway to be the more flexible one for video, audio, and multi-modal pipelines.

Quick comparison table

Dimension	Midjourney (V8.2, July 2026)	Runway (Gen-4.5 + GWM-1, July 2026)
Primary output	Still images, 5-second video clips	Generative video, real-time world simulation
Latest model	V8.2, released July 24, 2026	Gen-4.5, released December 1, 2025
Text-to-Video leaderboard	N/A	#1, 1,247 Elo (Artificial Analysis)
Image-aesthetic focus	Yes—personalization, moodboards, style refs	Secondary, also offers image models (Gen-4 Image, Nano Banana Pro)
Real-time interactive media	No	Yes—GWM-1, GWM Worlds, GWM Avatars
Robotics & simulation	No	Yes—GWM Robotics SDK
Hollywood case study	Limited public references	House of David, Lionsgate equity deal (June 11, 2026)
Asset ownership	Yes, with a $1M revenue gate requiring Pro/Mega	Yes, per Terms of Use
Privacy / Stealth	Stealth mode on Pro/Mega	Per workspace and plan defaults
Pricing (annual)	Tiered subscriptions, $1M revenue rule on ownership	Free / $12 / $28 / $76 per month on annual plans
Team size	~60 (self-funded research lab)	Offices in NYC, SF, Seattle, London, Tel Aviv, Tokyo
Brand scale	Midjourney.com homepage + Discord	“60M+ creators” (runwayml.com)
Hardware	Hardware livestream announced June 17, 2026	GWM Robotics SDK, NVIDIA Hopper + Blackwell

What each tool does best in 2026

Where Midjourney still wins

Midjourney wins when the job is still imagery, mood, and aesthetic exploration. Concept art, brand moodboards, editorial-style stills, and the early creative direction phase of a campaign. V8.2 sharpens the bet with smarter personalization and a more “edgy” output profile. If your pipeline is Figma, brand systems, and static social channels, Midjourney is the more direct evaluation.

Where Runway clearly wins

Runway wins when the deliverable moves, speaks, or simulates. Text-to-video, image-to-video, real-time conversational avatars, robotics policy evaluation, and AI-assisted VFX. Gen-4.5 leads the Artificial Analysis leaderboard. GWM-1 puts Runway in a category Midjourney doesn’t touch. The Lionsgate equity deal and the House of David case study give Runway a real production pedigree that Midjourney has not matched in public.

Where the lines are blurring

Two things are worth flagging.

First, Runway is now a multi-model platform. Since February 20, 2026, its Creative suite has included Kling 3.0, Kling 2.6 Pro, Kling 2.5 Turbo Pro, WAN2.2 Animate, GPT-Image-1.5, and Sora 2 Pro alongside Runway’s own models (runwayml.com/changelog). You can generate Veo 3.1, Nano Banana Pro, Lyria 3, and others from the same workspace. Midjourney does not aggregate third-party models.

Second, Midjourney is diversifying. Midjourney Medical is a real, funded side project aiming for 50,000 Ultrasonic CT scanners by 2031 (midjourney.com/medical). The Co-Star acquisition brings a consumer-app team of about two dozen people in-house (updates.midjourney.com/midjourneys-first-acquisition; TechCrunch, July 24, 2026). These moves are not directly comparable to Runway’s product surface—but they tell you Midjourney’s research lab is not betting its future on image generation alone.

Who should pick what

1. Art directors and brand designers — start with Midjourney V8.2. Personalization, style references, and the moodboard workflow are purpose-built for this.
2. Filmmakers and motion marketers — shortlist Runway Gen-4.5 and Aleph 2.0. Add GWM-1 if you also need real-time simulation or world-building.
3. Social content teams — pick by format mix. Stills-first social maps to Midjourney; motion-first short-form maps to Runway.
4. Production studios — treat both as previsualization tools. For end-to-end generative pipelines, the Lionsgate co-development and House of David workflow are the most credible production references in public (runwayml.com/news/runway-and-lionsgate-expand-partnership; runwayml.com/customers/how-house-of-david-used-runway-to-become-amazons-latest-hit-series).
5. Robotics and simulation teams — Runway GWM-1 Robotics SDK is the only mainstream creative-AI surface that targets this workflow. Midjourney has no equivalent.
6. Agencies over $1M annual revenue — factor in the Midjourney ownership gate. You’ll need Pro or Mega to own what you generate, on top of your normal usage fees (docs.midjourney.com/hc/en-us/articles/32083055291277-Terms-of-Service).

Limitations both tools still have

• Runway Gen-4.5 still has known limits: causal reasoning can fail (a door opens before the handle is pressed), object permanence is imperfect (occluded items can disappear), and there’s a “success bias” where actions disproportionately succeed (runwayml.com/research/introducing-runway-gen-4.5).
• Midjourney V8.2 is described as “more creative, bold, sophisticated, edgy and fresh,” but it is still subject to the same content policies as prior versions—no adult content, no political campaigns, no deceptive content (docs.midjourney.com/hc/en-us/articles/32083055291277-Terms-of-Service).
• Both still require a human finishing pass for brand-safe, on-system work. Neither replaces Figma, After Effects, DaVinci Resolve, or a final human review.

Bottom line

Choose Midjourney for still-image creative direction under V8.2. Choose Runway for generative video, real-time world simulation, and motion production under Gen-4.5 and GWM-1. The decision almost always comes down to whether your next deliverable is a still or a sequence—not which model is louder on social this week.

If you need both, budget for both. The two products are now too specialized to ask either one to do the other’s job in 2026.