Full-Time

QA Analyst

Midjourney

Midjourney

201-500 employees

AI text-to-image generation via Discord bot

Compensation Overview

$100k - $125k/yr

San Francisco, CA, USA + 1 more

More locations: New York, NY, USA

Hybrid

Hybrid work is required in San Francisco or New York.

Category
QA & Testing (1)
Required Skills
Software Testing
Quality Assurance (QA)

Get referred to Midjourney

See people who can refer or advise you

Requirements
  • Several years of experience testing software used by real people.
  • Ability to understand how products behave, investigate subtle bugs, and reason through complete user flows.
  • Ability to collaborate effectively with designers, engineers, and product managers.
  • Commitment to quality and testing throughout development rather than only before release.
  • Experience using tools such as Claude Code and Cursor.
Responsibilities
  • Partner closely with product, design, and engineering throughout the development process.
  • Lead testing efforts on the most important product initiatives.
  • Investigate difficult bugs, identify edge cases, improve test coverage, and help teams understand product behavior in the real world.
  • Design test plans and identify issues before they reach users.
  • Improve software testing through better tooling, automation, and testing practices.
  • Reproduce issues, identify root causes, and verify fixes before release.
Desired Qualifications
  • Interest in products without established playbooks.
  • Enjoyment of understanding how products behave, not just whether they work.

Midjourney operates as an independent research lab offering an AI image-generation service. It uses a proprietary text-to-image model accessed through a Discord bot, where users can input textual prompts to generate, refine, and upscale digital art. The service is in open beta and runs on a subscription model with multiple tiers that determine access and features. The company is self-funded and does not rely on venture capital, appealing to artists, designers, and creative enthusiasts who want to explore AI-generated art and creative tooling. Unlike many competitors, it emphasizes a self-funded, Discord-based user experience with a focus on community and iterative refinement of images through prompts and upscaling. Its goal is to provide accessible, scalable AI-powered image creation to a wide range of creative professionals and hobbyists while advancing the field of generative art.

Company Size

201-500

Company Stage

N/A

Total Funding

N/A

Headquarters

San Francisco, California

Founded

2021

Get referred to Midjourney

See people who can refer or advise you

Simplify Jobs

Simplify's Take

What believers are saying

  • V8.2 edit testing on August 27, 2026 expands production use for creative teams.
  • Co-Star’s 4.3 million monthly users give Midjourney immediate consumer distribution.
  • Banu Guler as chief design officer strengthens onboarding, retention, and standalone-app execution.

What critics are saying

  • Disney, Universal, and Warner Bros. can pursue damages that threaten Midjourney’s core business.
  • Discord dependence still limits distribution; a delayed standalone app cedes users to OpenAI and Adobe.
  • Co-Star’s young audience may reject AI branding, causing churn after the July 2026 acquisition.

What makes Midjourney unique

  • Midjourney’s July 2026 Co-Star acquisition adds consumer design and mobile product talent.
  • The August 27, 2026 V8.2 edit model centers controlled revision, not just generation.
  • Its independent, bootstrapped structure lets Midjourney move without investor pressure.

Help us improve and share your feedback! Did you find this helpful?

Benefits

Hybrid Work Options

Remote Work Options

Growth & Insights and Company News

Headcount

6 month growth

-4%

1 year growth

-4%

2 year growth

-3%
Geeky Gadgets
Aug 31st, 2026
New Midjourney Edit Mode blends up to four reference images.

New Midjourney Edit Mode blends up to four reference images. Midjourney's latest update introduces the highly anticipated Edit Mode, a feature designed to enhance creative workflows through greater precision and interactivity. As highlighted by Future Tech Pilot, this mode brings several notable improvements, including the ability to blend up to four reference images seamlessly and attach visuals directly to prompts for more contextually accurate results. While the update offers exciting opportunities for digital artistry, some challenges remain, such as proportional inconsistencies in complex compositions and occasional unpredictability with intricate prompts. Explore how this update can reshape your approach to digital art with features like real-time editing for fine-tuning colors and objects, advanced image blending for unique compositions and customizable style transformations to align visuals with your creative goals. You'll also gain insight into practical strategies for maximizing Edit Mode's potential, from experimenting with blending techniques to refining prompts for optimal results. This guide provides the clarity you need to navigate the new features effectively and elevate your creative projects. Midjourney's new Edit Mode features. TL;DR Key Takeaways: * Midjourney's new Edit Mode introduces advanced tools like real-time editing, image blending and seamless prompt integration, enhancing precision and interactivity in digital artistry. * Customization features include support for blending up to four reference images, background and style specification and applying artistic transformations like anime or abstract effects. * Performance improvements focus on fast processing and creative combinations, allowing efficient workflows and innovative outputs. * Limitations include occasional proportional inaccuracies, inconsistent results with complex prompts and gaps in style transformation effectiveness. * The alpha.Midjourney.com platform update offers an enhanced interface and streamlined navigation, improving usability for both new and experienced users. The Edit Mode introduces a variety of tools aimed at improving your editing experience and expanding creative possibilities. Some of the standout features include: * Attach to Prompt: This tool allows you to integrate images directly into your prompts, using visual references to guide the AI for more accurate and contextually relevant results. * Quick Edit Tool: Make real-time adjustments to images, such as tweaking colors, modifying objects, or refining styles, giving you greater control over your creative process. * Image Blending: Combine multiple reference images to create cohesive and unique compositions, allowing you to explore new artistic directions. These tools are designed to streamline workflows, allowing you to achieve your desired outcomes with greater ease and precision. By combining intuitive functionality with advanced technology, Edit Mode enables you to bring your creative visions to life. Enhanced customization options. Customization lies at the heart of this update, offering tools that allow you to tailor your creations to your specific needs. Key enhancements include: * Support for Multiple References: Blend up to four reference images seamlessly, allowing you to incorporate diverse elements into a single composition. * Background and Style Specification: Define backgrounds and artistic styles to create visuals that align with your unique vision. * Style Transformations: Apply artistic aesthetics, such as anime or other stylized effects, to suit your creative goals, whether they lean toward realism or abstraction. These features provide a robust framework for experimentation, allowing you to refine your work and explore new possibilities in digital art. Whether you're creating for personal projects or professional purposes, the enhanced customization options ensure your results are both unique and impactful. Expand your understanding of Midjourney with additional resources from its extensive library of articles. Performance and efficiency. The new Edit Mode prioritizes performance and efficiency, making sure a smoother creative process that keeps pace with your ideas. Key improvements include: * Fast Processing: Enjoy quick turnaround times for edits and image generation, allowing you to iterate rapidly and maintain creative momentum. * Creative Combinations: Blend prompts and images to produce unique outputs that push the boundaries of digital art, offering endless opportunities for innovation. These advancements allow you to focus on the creative aspects of your work without being hindered by technical limitations. The streamlined process ensures that your time is spent on refining your vision rather than navigating cumbersome workflows. Limitations to consider. While the new Edit Mode offers significant advancements, it is not without its challenges. Some limitations to be aware of include: * Proportional Issues: Generated images may occasionally display inaccuracies in proportions or anatomy, particularly in complex compositions. * Inconsistent Results: Highly specific or intricate prompts can sometimes yield unpredictable outcomes, requiring additional adjustments to achieve the desired effect. * Style Transformation Gaps: Certain style changes may not produce noticeable differences, limiting their effectiveness in some scenarios. These challenges highlight areas where further refinement is needed to enhance the platform's reliability and consistency. Despite these limitations, the tools remain highly effective for a wide range of creative applications. Updates to the alpha website. To complement the new Edit Mode, the alpha.Midjourney.com platform has been updated with several improvements designed to enhance usability. Key updates include: * Enhanced Interface: A more intuitive interface simplifies workflows, making tasks like adding objects or text to images easier and more accessible. * Streamlined Navigation: Improved navigation reduces the learning curve, making sure that both new and experienced users can quickly adapt to the platform. These updates aim to create a more user-friendly experience, allowing you to focus on your creative projects without being distracted by technical hurdles. The improved interface ensures that the tools are accessible and efficient, regardless of your level of expertise. Tips for maximizing Edit Mode. To fully use the capabilities of the new Edit Mode, consider the following tips: * Be Specific: Provide detailed prompts and descriptions to guide the AI effectively, especially when specifying backgrounds, styles, or other key elements. * Experiment with Blending: Combine different reference images or use various style codes to discover unique combinations and creative possibilities. * Iterate and Refine: Don't hesitate to make multiple adjustments to your prompts or edits to achieve the best possible results. By adopting these strategies, you can unlock the full potential of the platform and create visuals that align with your artistic vision. The flexibility and precision of Edit Mode make it an invaluable tool for both novice and experienced digital artists. Looking ahead. Midjourney's new Edit Mode represents a significant step forward in image editing technology. With its focus on speed, flexibility and innovation, the update provides a powerful toolset for creating dynamic and customized visuals. While some limitations persist, the expanded features and improved interface make it an essential resource for digital artists and creators. As the platform continues to evolve, it promises even greater opportunities for creative exploration and expression, solidifying its role as a leader in the digital art space. Disclosure: Some of its articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about its Disclosure Policy.

SynthMinds
Aug 31st, 2026
Midjourney's new V8.2 edit model changes the creative workflow, not just the image.

Midjourney's new V8.2 edit model changes the creative workflow, not just the image. Midjourney opened broader testing of its first V8.2 image edit model on 27 August 2026. Two days later, it issued a further quality update to the model. The release sits on top of V8.2, which became Midjourney's current default image model in July. The important change for brand and creative teams is not another increment in image quality. Midjourney is moving more of the workflow from one-shot generation into controlled revision: edit an existing image with instructions, use several images as references, change a selected area, expand a canvas and carry existing style systems into the edit process. That makes the tool more useful for production. It also moves governance deeper into the workflow. What Midjourney released. Midjourney says the V8.2 edit model supports instruction-based image editing, generation from up to four image references, inpainting, outpainting, personalization, moodboards and style references. The company describes the release as an early model that it is opening to a larger testing community. Its 29 August update specifically says image quality in the edit model was improved following early feedback. That is useful context for teams evaluating it: this is a live, fast-changing production surface, not a frozen specification. V8.2 itself remains focused on aesthetics, image quality and personalization. Midjourney says the model should produce more creative and sophisticated images, reduce random low-quality results and better understand users' aesthetic preferences. Those are vendor claims, not a guarantee that the model will behave consistently on a particular brand system. Editing changes where quality control happens. In a prompt-only workflow, the common pattern is generate, reject, regenerate. Editing introduces another possibility: keep the useful parts of an image and change only what is wrong. That sounds incremental, but it changes the economics of creative iteration. A team can preserve composition, character, art direction or product placement while changing a local detail instead of starting from zero. Outpainting can extend an approved image for another format. Multiple references can help anchor the output to an established visual world. The practical question therefore shifts from "can the model make a good image?" to "can the team revise an image without losing what has already been approved?" That is a more production-oriented test. References become part of the source record. The new workflow also increases the importance of what is supplied to the model. If a user can attach up to four image references and ask Midjourney to generate or edit against them, reference selection becomes part of the creative decision. Teams should know whether those images are owned, licensed, commissioned, internally created or merely found online. A tool's ability to accept a reference does not establish permission to use it. Nor does a successful generation establish that the resulting asset is clear of third-party rights or suitable for every channel. For important work, the source record should therefore cover both the final output and the material used to steer it. What brand teams should test before adopting it. A useful pilot does not need to ask whether V8.2 is "better" in the abstract. Test the jobs that repeatedly create friction in your own workflow. Can an approved key visual be reformatted without changing the subject? Can product details survive a local edit? Can a campaign character remain recognisable while wardrobe or environment changes? Can the same art direction be extended across landscape, square and vertical formats? Can an unwanted object be removed without introducing another problem elsewhere? Then test the governance around those actions. Who is allowed to upload reference material? Which assets may contain confidential or pre-launch information? What source records are retained? Which changes require another brand, legal or client approval? How is the final approved version distinguished from intermediate generations? Those controls are less exciting than the model demo. They determine whether the capability can move beyond individual experimentation. The larger shift is from generation to revision. Generative image tools first won attention by creating something from nothing. Production teams often need something less theatrical: change this, preserve that, extend this frame, keep the character, use this style system, fix only this area. Midjourney's V8.2 edit model is another sign that the competitive surface is moving towards controllability and revision rather than generation alone. For agencies and in-house teams, that may matter more than a marginal jump in aesthetic quality. A model that fits an approval-and-revision workflow can remove more real production effort than one that simply makes a stronger first draft. What synthminds is watching. The model is still changing quickly, so the useful evidence will come from repeated production tests rather than launch examples. Synthminds Branding And is watching how reliably V8.2 preserves approved elements across edits, how reference-heavy workflows behave, and whether the model can reduce regeneration without creating a new layer of review work. The immediate recommendation is simple: test it as an editing system, not as another image generator.

goongen.ai
Aug 28th, 2026
Genshin Impact AI art: generate Yelan, beidou and more (2026 guide).

Genshin Impact AI art: generate Yelan, beidou and more (2026 guide). Aug 28, 2026 anime characters Genshin Impact has one of the largest fan art communities of any game in the world, and AI generation has quietly become a big part of it. The problem: type "Yelan from Genshin Impact" into a generic AI image generator and you get a blue-haired woman who is almost, but not quite, entirely unlike Yelan. Wrong outfit, wrong eyes, missing the mole, hair parted the wrong way. Fans notice every detail, because the details are the character. This guide covers how Genshin Impact AI art actually gets consistent - character LoRAs on an anime-native model - plus danbooru-tag prompt tips and the adult members of the cast you can generate right now. Quick disclosure up front: Goongen built goongen.ai, one of the tools discussed below, so weigh its opinion accordingly. Goongen will keep the competitor section honest. Why generic generators get Genshin characters wrong. General-purpose models like DALL-E, Firefly or Flux are trained mostly on photos and broad web data. They know Genshin Impact exists, and they can approximate a famous character's silhouette, but they have no reliable internal reference for the exact costume design, accessory placement, or color palette. Every generation re-rolls the dice on those details. There is a second, growing problem: mainstream tools are becoming cautious about named characters at all. Midjourney is currently being sued by Disney, Universal and Warner Bros. specifically over its ability to output copyrighted characters, and that case was still in discovery as of this writing. Whatever the outcome, the pressure pushes big platforms toward blocking or degrading character-name prompts. Anime-focused tools built on open models do not have that constraint baked in. The technical fix for consistency has existed for years in the Stable Diffusion community, and it is two layers deep: * An anime-native base model. The Illustrious / SDXL family is trained on the danbooru imageboard dataset, which contains an enormous amount of tagged Genshin fan art. The model already knows character tags, outfits, poses and styles natively. * A character LoRA on top. A LoRA is a small fine-tune trained on one character. It locks in the exact design - Yelan's asymmetric hairstyle, Beidou's eyepatch, Xinyan's spiky red-streaked hair - so the character stays on-model across hundreds of generations instead of drifting. Base model for style and understanding, LoRA for identity. That combination is why dedicated anime tools produce recognizable Genshin characters and generic tools produce cosplay-adjacent strangers. If you want the deeper background on the model family, its anime AI character creator guide goes into detail. Prompting with danbooru tags. Because Illustrious models are trained on danbooru data, they respond best to danbooru-style tags rather than long natural-language sentences. A solid Genshin prompt looks like this: yelan (genshin impact), 1girl, solo, mole under eye, smirk, night, city rooftop, neon lights, looking at viewer, dynamic pose, masterpiece, best quality A few tips that consistently improve results: * Lead with the character tag. Use the danbooru form, series included: beidou (genshin impact), ningguang (genshin impact). The earlier it appears, the more weight it carries. * Do not over-describe the default outfit. The LoRA already knows it. Listing every belt and tassel just gives the model chances to contradict itself. Describe clothing only when you want something different - casual, swimsuit, office lady - and the LoRA keeps the face and hair on-model while the outfit changes. * Use real booru tags for scenes and poses. Tags like sitting, from behind, cowboy shot, sunset, cherry blossoms are all vocabulary the model has seen hundreds of thousands of times. Made-up phrases work worse than boring standard tags. * One character per image. Multi-character prompts cause trait bleeding - Beidou's eyepatch migrating onto Ningguang, hair colors mixing. Generate separately and composite if you need a group shot. * Keep a quality suffix. masterpiece, best quality plus a negative prompt with lowres, bad hands, bad anatomy is still the boring, effective standard. If you would rather not memorize tag vocabulary, goongen.ai's anime mode has a click-together prompt builder - pick the character, then click scene, pose and outfit chips and it assembles the tags for you. More prompt patterns live in its prompt library. Genshin characters on goongen.ai. goongen.ai's anime mode runs a WAI Illustrious SDXL model with a library of over 1,400 character LoRAs, including roughly 70 from Genshin Impact. One deliberate policy note: this post features only clearly adult characters, and the platform is for fictional adult characters - no real people, ever. Here are standouts from the roster, with example galleries on each page: | Character | Region | Why she generates well | | Yelan | Liyue | Striking asymmetric design, the mole and smirk land reliably | | Beidou | Liyue | Eyepatch, scar and captain's confidence - very hard to mistake | | Xinyan | Liyue | Rock-star energy, great for concert and stage scenes | | Ningguang | Liyue | Elegant, ornate outfit that the LoRA holds together | | Ganyu | Liyue | The horns and gradient hair stay consistent across poses | | Shenhe | Liyue | Silver braid and bodysuit, strong in cold color palettes | | Lisa | Mondstadt | Witch-hat librarian, works in both cozy and dramatic scenes | | Jean | Mondstadt | Knightly poses and armor variants come out clean | | Eula | Mondstadt | Dancer-knight poses, excellent with dynamic motion tags | | Rosaria | Mondstadt | Gothic silhouette, thrives in night and church settings | | Raiden Shogun | Inazuma | One of the most-trained characters in the danbooru dataset | | Yae Miko | Inazuma | Fox ears and shrine-maiden outfit, playful expressions | | Dehya | Sumeru | Flame-maned mercenary, great for action compositions | Beyond the model, the pitch is the infrastructure: no email verification, outputs locked so that only your password can open them (the server cannot read what you make, so your gallery stays your business - there is no one here who could browse or judge it), credit-based pricing with crypto accepted, no subscription. That premium setup is on purpose - credits are the only revenue stream, which means no ads, no email spam, and no data being mined or sold behind the scenes. There is an Endless mode that turns any character into an infinite generation feed, and the free demo gives you 5 images with no card and no signup friction, so you can check whether your favorite character holds up before spending anything. Where else you can make Genshin AI art. Fair is fair - goongen.ai is not the only option, and depending on what you want, it may not be the best one for you. | Tool | Genshin support | Adult content | Account required | | goongen.ai | ~70 built-in LoRAs | Uncensored | No email verification | | SeaArt | Community LoRAs | Age-verified users, as of this writing | Yes | | PixAI | Community LoRAs | Gated 18+ | Yes | | Tensor.Art | Community LoRAs | Prohibited, as of this writing | Yes | | Midjourney | No LoRA system | No | Yes | | Local ComfyUI | Civitai LoRAs | Unrestricted (your hardware) | None | SeaArt and PixAI are genuinely good anime platforms with huge community LoRA libraries - almost every Genshin character exists there in multiple versions. The tradeoffs are accounts, cloud-stored galleries, and moderation layers that shift over time. Its NovelAI vs PixAI vs SeaArt comparison covers them in depth. Tensor.Art used to be in the same bucket but moved to prohibit NSFW generation - the full story is in its Tensor.Art NSFW policy post. For SFW Genshin fan art it still works fine. Local ComfyUI with LoRAs downloaded from Civitai is the maximum-control option: free, private, unrestricted. The cost is a capable GPU and real setup time. If that sounds like you, see its ComfyUI alternative post for the honest tradeoff discussion. Genshin Impact and all its characters belong to HoYoverse (miHoYo). HoYoverse has historically been friendly to fan creations - its published derivative-work guidelines permit fan illustrations, doujinshi and even small merchandise runs under specific conditions - but those guidelines predate the AI art wave and do not explicitly address generated content, as of this writing. The sensible approach: treat AI Genshin art as personal fan art. Enjoy it, share it labeled as fan-made if you post it, and do not sell it. goongen.ai is not affiliated with, endorsed by, or connected to HoYoverse in any way. The bottom line. Consistent Genshin Impact AI art is a solved problem, but only on the anime-native stack: an Illustrious-family model plus a character LoRA plus danbooru-tag prompting. Generic generators will keep giving you almost-Yelans, and mainstream platforms are trending toward blocking character names entirely while the lawsuits play out. Pick based on what you value: community features (SeaArt, PixAI), full local control (ComfyUI), or privacy without setup - which is the specific niche goongen.ai occupies. The demo is 5 free images with no card, so the cheapest way to decide is to just generate Yelan and see if it is actually her.

Need AI Tool
Aug 21st, 2026
Midjourney rolls out V8.2 & Multi-Reference 'oref' consistency Engine.

Midjourney rolls out V8.2 & Multi-Reference 'oref' consistency Engine. Next-gen multi-angle character locking and unified visual consistency for generative creators. Ethan Walker August 21, 2026 ~380 words Key takeaways. * Multi-Reference 'oref' Engine: Pass up to 4 distinct reference images simultaneously to lock facial features, outfits, and color palettes. * V8.2 Aesthetic Fidelity: Substantial photorealism leap with refined skin micro-textures, specular highlights, and natural hand anatomy. * Web Canvas Integration: Integrated directly into the alpha Midjourney web creation suite for real-time inpainting and region locking. The character consistency breakthrough. Generating persistent characters across multiple scenes has long been the primary bottleneck for AI storyboarding, graphic novels, and game concept art. With the release of Midjourney V8.2, the research team has introduced an updated attention-routing mechanism dubbed -oref (Omni Reference). Unlike earlier `-cref` implementations which often drifted in complex lighting scenarios, `-oref` computes cross-attention matrices against front, 3/4, and profile angle images simultaneously. The result is near-identical character fidelity even when transitioning between dramatic close-ups and wide action sequences. Ecosystem comparison & availability. Midjourney V8.2 is live today for Pro and Mega tier subscribers via Discord and the Web suite. Creative directors looking for alternatives can explore Midjourney, Flux.1, and Recraft on NeedAITool. Found this useful? Share it: Ethan Walker I'm a technology writer passionate about AI tools, automation, productivity software, and emerging SaaS platforms. I spend my time testing digital tools and breaking down complex technologies into practical insights that help businesses, creators, and professionals work smarter.

Breakit
Aug 18th, 2026
Daniel Ek's health company Neko Health opens at the luxurious address.

Daniel Ek's health company Neko Health opens at the luxurious address. Now it becomes reality. The health scanning startup Neko Health, founded by Spotify co-founder Daniel Ek, opens its first American office on September 24 - in the middle of the Soho district in Manhattan. The office will be located at 300 Lafayette Street in Soho, Manhattan's luxury shopping district, and the customer list is impressive - more than 25,000 New Yorkers have already signed up for the waiting list, the company states. TechCrunch writes. Neko Health was founded by Spotify CEO Daniel Ek and Hjalmar Nilsonne, who is the company's CEO. The company has developed its own proprietary technology for body scanning that is combined with blood tests and data from exercise devices to assess a person's health. The startup already has operations in the United Kingdom and Sweden. Last month, Neko Health raised $700 million in a new funding round. In connection with that, the company stated that 100,000 people had already undergone scans. Neko is one of several health scanning companies founded by tech entrepreneurs or financed by venture capital firms. The AI lab Midjourney is also reported to be developing a body scanner that will be integrated into a spa experience, with a planned opening in San Francisco in 2027. Tony Robbins and Peter Diamandis run the company Fountain Life, which performs scans and health assessments. Function Health, founded by Dr. Mark Hyman, offers blood tests and has expanded to include body scans after the acquisition of startup Ezra. This article is written with the help of the AI tool Ahody.