Midjourney v8

• 1. overview (657 chars)

  Midjourney v8 is the new generation basic model of the image generation platform Midjourney since v7 in 2025. The first v8 Alpha was opened for community testing on March 17, 2026, and v8.1 Alpha followed on April 14, and officially became the default model in June 2026, replacing v7. The core upgrades of this generation focus on three things: the generation speed is about five times faster, the native 2K resolution no longer relies on amplification, and the ability to follow text and prompt words has improved significantly. It is still recognized as the "best-looking" image model, but text rendering is still shortcomings, and because it is closed source, has no API, and lacks legal exemption clauses, it is not the safest choice in compliance-sensitive commercial projects.

• 2. Product background (837 chars)

  Midjourney was founded by David Holz, who also co-founded Leap Motion. The company describes itself as a "community-funded research laboratory" with a team of about 60 people, and its positioning is to "build the most beautiful AI models in the world." Since v1 in 2022, Midjourney has been refreshing the ceiling of image aesthetics in almost every version. Community culture and the sharing generation mechanism of Discord/website are the foundation of its network effect. In June 2026, the company also previewed its first hardware project, saying that there were four hardware directions under development, and its ambitions were no longer limited to software.   The official definition of v8 is not an "improved version of v7", but a new code base rebuilt from scratch. v8 Alpha was first in grayscale at alpha.midjourney.com, v8.1 Alpha was later released, and v8.1 became the default model for the main site and Discord. v8.0 Alpha was deprecated about two weeks after v8.1 became official.

• 3. Product functions and usage experience (1967 chars)

  Speed ​​is the most intuitive change. Officials say that standard tasks are about four to five times faster than v7. A third party has actually measured the same prompt word in v7 in about 45 seconds and v8 in less than 8 seconds; v8.1 takes about 4 seconds to produce a picture in SD (standard resolution) and about 12 seconds in HD. Iterative feedback is compressed from "waiting by minutes" to "waiting by seconds", which has a great impact on exploratory creation.   The new --hd parameter allows images to be rendered directly at native 2K, eliminating the need for a separate upscaling step and keeping details in place from the start. Text rendering is the most obvious point of progress in this generation: put the text in the "double quotes" of the prompt word, v8 can render readable signs, labels, and poster titles. Although it is still not perfect, it is a qualitative leap compared to v7. However, it should be noted that in multiple horizontal evaluations in mid-2026, Midjourney’s text capabilities are still the weakest among head tools, and Ideogram, GPT Image 2, and Google Nano Banana are all more reliable.   v8 is more literal in the matter of "obeying". v7 will proactively add a layer of creative interpretation to your prompts. v8 is closer to the original words you wrote, which is very valuable for product pictures, architectural renderings, commercial images, and pictures with text; but the price is that it lacks the warm, painterly, and atmospheric "artistic bias" of v7. Some experienced users bluntly said that v8 is "more realistic, but less artistic and less surprising." v8 also brings improved conversation mode, grid view, sidebar settings, and is compatible with v7's personalized archives, mood boards, and style references (sref). The official recommendation is to adjust --stylize to 1000 to rely heavily on the style system.   The stability of characters and anatomical structures has been significantly improved, the proportions of hands, faces, and bodies are more consistent, and physical lighting (reflections, shadows, material interactions) are more like the real world. However, v8 still has an obvious "alpha temper" - founder Holz himself described it as "smart, but stupid in some unexpected places", and it will still overturn when following complex relationships and rare subject combinations.

• 4. Pricing and business model (1010 chars)

  v8 is included with users' existing Midjourney subscriptions, and the subscription tiers themselves remain unchanged: Basic at $10 per month (~3.3 fast GPU hours, no Relax, no incognito), Standard at $30 per month (15 hours + unlimited Relax), Pro at $60 per month (30 hours + stealth mode + 12 concurrency), and Mega at $120 per month (60 hours). Annual payment saves 20%, additional GPU time is about $4/hour, and all paid plans include commercial usage rights. Companies with more than $1 million in annual revenue are required to use Pro or Mega.   The real cost trap is not in the subscription itself, but in the advanced functions of v8: --hd, --q 4, style reference, and mood board. Each generation time and cost is four times that of the standard version, and the superposition can reach sixteen times; and v8 does not support Relax mode (only Fast) in the initial stage of launch, and each generation consumes fast GPU hours. For heavy users, this means budget pressure is focused on these switches rather than the model itself. Midjourney does not have a free trial, Basic is the only way to get started, and there is no official API. Programmatic calls can only be made through Discord or unofficial channels.

• 5. User reviews (766 chars)

  Positive feedback focused on speed and controllability: one user said, "What used to take 30 to 60 seconds is now done in 10 seconds. The difference is huge when I change the prompt for the 30th time." Product photography, architectural drawings, and signboard pictures with text are considered "finally available", and the consistency of characters is also more stable than in the past. Negative feedback mainly comes from two types of people: one is old users, who have spent months rating and polishing the v7 personalized profile to make v7 "just right". Switching to v8 is like comparing a new model to an old tool that understands their own tastes, and they naturally feel inferior; the other is those who pursue the ultimate artistic sense, who think v8 is colder, more neutral, and less surprising. Most of the complaints about "v8 is worse than v7" are actually based on familiarity rather than quality itself.

• 6. Industry evaluation (663 chars)

  In the image model landscape of 2026, Midjourney v8 is recognized as the "King of Aesthetics", but the overall champion has changed hands. GPT Image 2 (OpenAI, April 2026) tops the list of most reviews with about 99% text accuracy, 4K output and conversational editing; Google's Nano Banana 2/Pro is leading in photorealism and portraits and is free and available; FLUX.2 is the best open source weight that can be self-hosted; Ideogram 4.0 specializes in typesetting. Midjourney is absent from most rankings because it is closed source, has no API, and does not appear in the public arena, but it still ranks first in the creators' aesthetic vote. In a word: choose it if it has a "sense of design", and choose a competing product if it is "obedient, able to write, and can be embedded in workflow".

• 7. Controversies and Risks (1237 chars)

  Copyright is the biggest problem for Midjourney. In June 2025, Disney, Universal, DreamWorks, Warner, DC, Hanna-Barbera and others jointly sued Midjourney, calling it a "typical copyright free rider" who can generate highly recognizable Yoda, Darth Vader, and Marvel characters without special prompts. They claimed US$150,000 per work and requested an injunction or even a temporary shutdown of the service; the cases have been merged, and mediation is required to be completed before August 2026. The court hearing is scheduled for By the end of 2026, potential damages amount to billions of dollars. The infringement accusation against Midjourney in the artist class action lawsuit (Andersen v. Stability AI, etc.) was once dismissed (the judge believed that the "compressed copy" theory was unclear), but was allowed to be reasserted after correction. Article 53 of the EU AI Act requires general AI providers to publish training data summaries, which will be fully implemented in August 2026. Non-compliance fines can be up to 15 million euros or 3% of global revenue.   In addition, closed source brings "zero disclaimer" - you own the output, but the platform does not cover copyright disputes; images generated by the Basic/Standard plan are published in the gallery by default, and competing products can see your prompts and finished products; video generation has not been updated for a long time, and it has lagged behind compared to full-time video tools.

• 8. Applicable groups and usage suggestions (763 chars)

  Choose v8 (v8.1) when: you need accurate images of people, products, and buildings, write detailed and specific tips and hope to strictly implement them, the text in the picture is very important, or you need to use native 2K for final rendering. Choose v7 when: You want atmospheric, painterly, conceptual art and emotional expression, with short, impressionistic cues. A two-stage route for beginners: first use v7 to "rating" a large number of times to create a personalized file (target 200 times), then bring the file to v8 and add --p and --stylize 500 to start - this step can directly determine whether your picture "looks like you". Use standard resolution throughout the exploration phase, and only turn on --hd for images worth retaining, which can only be supported by a small GPU. If the work is mainly focused on precise typesetting and brand printing, Canva or Adobe are still more convenient tools.

• 9. Summarize (489 chars)

  Midjourney v8 achieves the three things of "fastest, most controllable, and most photo-like" at the same time. It is a generational leap in commercial imaging and realistic creation. However, it is still the "aesthetic professional" that is closed source, has no API, weak text, and is deeply involved in copyright lawsuits. Use it as an art director and don’t gamble on its immunity on compliance-sensitive client projects. Looking forward, v8.2 is already in early testing, and Relax mode and cheaper rendering files are also on the roadmap, which deserves continued attention.

No in-depth report yet

• 

  AllanCastro

  —

  V8.1 转正之后稳定多了，HD 模式下 2K 直出确实省了放大步骤，印刷级素材方便不少。

---

• 

  DrTamaraEdwards_dev

  —

  快是真的快，回不去了。

---

• 

  Jack220

  —

  免费试用早取消了，想试最少得掏 10 刀月费。

---

• 

  Dennis_Rogers52077

  —

  V8 出图几秒一张，迭代效率拉满，但默认风格感觉变平了。

---

• 

  JAfra

  —

  说实话 V8 不是全能模型，别神话它。它的价值很明确：够快、中文更强、大场景能打，最适合拉高画面审美上限。把它放在工作流最前面定视觉基调，比单张单张死磕连续剧情现实得多。真要做好用的生产后端，还是得靠更擅长编辑和一致性的工具去接力。

---

• 

  trueAdrianCaldwell_dev

  —

  中文文字这次终于能看了，做海报标题图实用性强了很多，记得把文字放进引号里再加 --raw。

---

• 

  DRcla

  —

  用了两周 V8.1，体感是审美还是顶级的，大场景、概念图、游戏原画那种气势确实能打。但人物一致性和连续分镜还是不够稳，想做整套角色剧情图最好别单独指望它。我现在把它当美术总监用，先定气质和氛围，后面连续资产交给 Nano Banana 2 接力，整个流程反而更快。

---

• 

  Natalie_Miller007

  —

  Trustpilot 上评分低主要不是图的问题，是退款政策和图片被删的投诉。图的质量没得说，客服和条款是真不友好。

---

• 

  LCastilloSr

  —

  老风格代码直接失效，养熟的习惯一夜作废，老用户真的会破防。

---

• 

  PaulWhite0078

  —

  Reddit 上有句话很形象：相机更精准了，但摄影师走了。V8 太听话反而少了点惊喜感，想要那味儿得开 --raw 或拉高 --stylize。

---

• 

  up_3sb564f9

  —

  贵，但香。

---

• 

  KyleEdwards87

  —

  成本这块得提醒新手：--hd 和 --q 4 都是标准任务的四倍 GPU 时间，两个一起开直接十六倍，Relax 模式早期还没法跟它们同时用。建议先用标准分辨率快速找方向，只对值得留下的图开高清精修，不然 GPU 小时眨眼就没了。

---

• 

  CDiaz168

  —

  国内直连基本没戏，得有合规海外网络环境才能用，普通用户门槛确实高。不过云端不吃本地显卡这点对没好卡的人挺友好。

---

• 

  KMillerII553

  —

  退款基本不可能，终身 GPU 用时超 20 分钟就不给退，订阅前想清楚。

---

• 

  Johnny_Ross_X

  —

  把 V8 和 Seedream 放一起看挺有意思：Seedream 更像生产工具，追的是理解和执行力；Midjourney 更像审美机器，守的是风格和质感。做海报、概念设计、带中文的视觉草图，MJ 依然很有竞争力，但一进高一致性角色和强可控商业生产，它就拉不开身位了。

---

• 

  Logan.Cooper

  —

  niji 还是那么顶，二次元插画那个角色感和色彩是 MJ 自己的灵魂，V8 再强也替代不了我对 niji 的偏爱。

---

• 

  BIwer

  —

  手还是偶尔翻车。

No reviews yet