By RamthaMedia
RamthaMedia Free eBooks · August 2026
Price: Priceless
· 11 min read
Preface
You type a sentence and an image appears – but that first image is rarely the one you actually pictured. Midjourney is built to close that gap: profiles that learn your taste, reference systems that lock in a face or a mood across dozens of images, a draft mode built for testing ideas fast, and an editor for reworking what you have already made. This book walks through how those pieces fit together – what to type, where to click, and why the same prompt gives two different people two different pictures.
Chapter 1
The Blank Prompt Bar Isn't as Blank as It Looks
You type four words into the prompt bar, wait a few seconds, and four images land in front of you – competent, on-topic, and not quite what you had in your head. That gap between the sentence you wrote and the picture you imagined is the entire reason most of what sits behind Midjourney's plain white prompt bar exists.
The model has to fill in everything you didn't specify – the lighting, the composition, the mood – and by default it fills those blanks with something close to an average of what the whole community tends to like. Midjourney has described this plainly: the algorithm's defaults are really the combined biases and preferences of everyone using it. The tool built specifically to change whose taste fills those blanks is called Personalization, invoked with –p.
Personalization learns from two things: pairs of images you rank against each other, and images you like on the Explore page. It doesn't need much to start working – a modest batch of ratings gets it moving, and it keeps sharpening for a while after that. You can check how many ratings you've logged from the personalize page at any point, and switch it on or off per prompt without losing the profile underneath.
None of this stays still. Midjourney has moved through a full run of major model versions in a short span, and each new one has changed what personalization, references and even basic editing can do – sometimes falling back to an older model for a feature that hasn't caught up yet. The real skill in using this tool isn't memorising the newest version number. It's knowing which of the many controls sitting around that prompt bar actually change what comes out, and which are still catching up to the model underneath them.
What you can actually do here
Midjourney has grown a long list of typed commands and sliders. The ones below are the ones that change what actually comes out.
Controlling what appears in the image
| Use | Who it fits | Where | Worth knowing |
|---|---|---|---|
| Teach the model your own taste | Anyone tired of generic default renders | Type –p after a prompt, or turn on Personalization in prompt bar settings | Rates images you like/dislike and shifts the model toward you Needs a batch of ratings before the effect is noticeable |
| Put a specific object, character or vehicle into a scene | Anyone building a recurring character or product shot | Drag an image into the Omni-Reference slot in the prompt bar, or use –oref url | Works on characters, objects, creatures, not just faces Competes with –stylize and –exp at high values |
| Keep a character's face while changing outfit or setting | Sequential art, story panels, recurring mascots | –cref url, with –cw 0 to focus on face only | cw 100 default carries face, hair and clothes; cw 0 carries just the face Built for Midjourney-made characters, not real photos of real people |
| Match the mood or style of a reference image | Consistent visual identity across a set | Drag into the Style Reference slot, or –sref url | Controlled by a weight slider, –sw Old sref codes can render differently after a version update – add –sv to pin the algorithm |
| Discover a style you didn't ask for and keep it | Anyone exploring rather than executing a fixed idea | –sref random, then save the code it generates | A shareable code takes you straight back to it Random means genuinely random – it will not repeat unless you save the code |
Controlling speed, quality and motion
| Use | Who it fits | Where | Worth knowing |
|---|---|---|---|
| Test an idea before committing to a full render | Early-stage exploration, permutations | Click the lightning-bolt icon in the prompt bar, or add –draft | Produces a batch of lower-resolution options at once Click Vary on the one you like to re-render it at full quality |
| Push a render into stranger, less expected territory | When the safe version of the prompt is boring | Add –weird after the prompt, alongside –stylize | Ranges from subtle to fully unpredictable Very high values reduce how closely it follows the actual prompt |
| Rerun an old standard-definition image at full resolution | Anyone who drafted first and liked the result | Click 'Rerun as HD' on the finished image | No need to retype the prompt or reference images |
| Turn a finished image into a moving clip | Short-form video, motion tests, social clips | Click Animate under an upscaled image; choose automatic or manual motion | Low motion suits ambient scenes; high motion suits full-scene movement High motion is more prone to visible mistakes |
| Extend a video clip beyond its first few seconds | Anyone whose clip cuts off too soon | Use the Extend option on a rendered video, up to four times | Each extension adds a further short segment There is a maximum number of extensions per clip |
| Reverse-engineer a prompt from an image you didn't write | Anyone trying to recreate or riff on a found image | Right-click an image and choose Describe, or drag it to the describe zone on web | Generates several candidate prompts from one image The prompts disappear once the page refreshes, so copy the one you want |
Chapter 2
Teaching the Model What You Find Beautiful
Two people can type the exact same sentence into Midjourney and get two different-looking results, once each of them has built a Personalization profile. The model isn't guessing differently for each person by accident – it has learned, from hundreds of small comparisons, what each of them tends to pick.
Building a profile is done from the Personalize page: you're shown pairs of images and asked which one you find more beautiful, over and over, or you simply like images as you browse Explore. Midjourney has rebuilt this interface more than once to make it faster and less like a chore, including a scrolling version that avoids the feeling of being asked to judge one image against another over and over.
You can hold more than one profile at a time – useful if you work across genuinely different projects, a moody photography series and a bright children's-book style, say – and name each one so you can switch between them. Moodboards work alongside profiles rather than instead of them: you build a board out of images whose mood you want, and a tightly-curated board produces focused results while a wide, varied one produces more unpredictable ones.
Both systems fold into an ordinary prompt with a code rather than a menu. Add –p after your text to use your active profile, and a moodboard's own code the same way. You can even blend more than one moodboard, or a moodboard with a style reference, in a single prompt – useful when a profile alone still isn't landing on the specific project you're working on.
Chapter 3
Putting the Same Face in Every Frame
A comic strip, a brand mascot, a recurring character in a set of story panels – all of them need the same face and the same outfit to survive across dozens of separate generations, and a plain text prompt has no memory of what it drew last time. Midjourney's answer to this moved through two systems before landing where it is now.
The earlier one, Character Reference, works by pointing –cref at an image of a character and controlling how strongly it's copied with –cw. At full strength it carries face, hair and clothes together; drop the weight toward zero and it focuses mainly on the face, which is the setting to use if you want the same person in a different outfit or setting. It was built with Midjourney-made characters in mind, not real photographs, and pushing it onto a real person's photo tends to distort them the same way any image prompt would.
Omni-Reference replaced most of what Character Reference did and extended it well past faces: it can hold a character, but also an object, a vehicle or a creature, with the instruction effectively reading as 'put THIS in my image.' The weight parameter is –ow, running from 0 up past the usual default of 100 into stronger territory, and it competes for influence with –stylize and –exp – so a heavily stylised prompt generally needs a correspondingly higher omni-weight to keep the reference visible at all.
There's one genuinely under-used move buried in how Omni-Reference was introduced: if your reference image itself contains two characters – either together in one frame or placed side by side across two images – and your prompt names both of them, it will often try to place both into the result. It's a rougher trick than the single-character case and needs a prompt that over-specifies which parts of each character to keep, but it's the closest thing this system has to a two-character scene generator.
You may also like:
What invideo Credits Actually Buy You
Chapter 4
Iterating at the Speed of Thought
Most of the cost of using Midjourney well isn't the credits – it's the time spent waiting on a render that turns out wrong. Draft mode exists entirely to shorten that loop: turn it on with the lightning-bolt icon in the prompt bar, or add –draft, and instead of one careful render you get a batch of quicker, lower-resolution options to react to.
The point of a draft isn't to keep it – it's to find the one worth finishing. Click Vary on whichever draft is closest to right and it re-renders at full resolution and quality. Because draft jobs render several options for roughly the cost of one standard job, this is consistently the fastest way to explore a prompt's range before committing.
Once you've settled on a direction, a handful of parameters shape how far the final render strays from the safe, expected version of your prompt. –stylize turns Midjourney's own aesthetic instincts up or down. –weird pushes toward genuinely unusual compositions, and at high values will noticeably loosen how closely the image follows your actual wording. –exp is newer and works a little differently – it tends to add detail and a more dynamic, tone-mapped feel, and combines with stylize rather than replacing it, though very high combined values will erode prompt accuracy the same way weird does.
None of these interact cleanly with each other at extremes. A prompt running high stylize, high exp and an omni-reference at once needs its reference weight raised specifically to compete – a detail easy to miss if you're only tuning one slider at a time and wondering why the character keeps drifting.
Chapter 5
When the First Draft Isn't Right
Not every fix belongs in the prompt. Sometimes the composition is right and only one object in the frame is wrong, and re-rolling the whole image to change one thing wastes everything that was already working. That's the job of Midjourney's external editor, which is available across every membership tier and reachable through the Edit button in either the sidebar or the lightbox.
Inside it, reframe, repaint, vary-region, pan and zoom all live in a single interface rather than as separate tools you have to remember the names of. A smart-selection feature lets you click directly on an object to mark it for removal or replacement, and you can import more than one image onto layers to build a collage or feed several source images into one retexture pass.
Retexturing is worth understanding on its own terms: rather than redrawing the scene, it estimates the shape of what's already there and reworks the lighting, materials and surfaces over that shape – useful for turning a rough block-out into something with a finished look without disturbing the composition underneath it.
All of this is controlled through text prompting and region selection together, and it works alongside personalization, style references, character references and image prompting rather than replacing any of them. If you're doing careful region-level work – marking exactly where a repaint should stop, or painting a precise mask for retexturing – a mouse gives you far less control than a pressure-sensitive pen does, which is the one place in this whole workflow where the input device itself starts to matter.
Chapter 6
From Still Image to Moving Scene
Midjourney's video system starts from a finished image rather than from a blank prompt – the workflow is called Image-to-Video, and it's reached by pressing Animate on any image you've already made. From there you choose between an automatic motion prompt, where the system invents how things should move, and a manual one, where you describe the motion yourself.
Two settings decide how much actually happens on screen. Low motion keeps the camera mostly still and lets the subject move in a slow, deliberate way – the risk is a clip that barely moves at all. High motion sets both camera and subject moving, which reads as more dynamic but is also more prone to visible mistakes as the model tries to keep everything coherent at once.
A finished clip doesn't have to stay short. Extend lets you add roughly four more seconds at a time, repeatable up to four times, and turning on remix mode during an extension lets you shift the prompt partway through rather than continuing the exact same motion. Looping and a defined end frame are both supported too – use –loop for a clip that repeats cleanly, or –end with a second image to tell the system where the motion should land.
You can also start from an image that was never made on Midjourney at all: drag it into the prompt bar, mark it as the start frame, and describe the motion the same way you would for a native image. HD rendering exists for video as well, at native 720p, but it costs several times more than a standard render and is aimed at people who genuinely need the extra clarity for professional delivery rather than for casual exploration – it's also restricted to the higher membership tiers and unavailable in relax mode.
You may also like:
Leonardo Turns One Login Into an Entire Creative Studio
Chapter 7
Building a Style Once, Reusing It Everywhere
A brand's visual identity, a book series' illustration style, a photographer's signature look – all of these need to survive across many separate prompts, and Style Reference is the system built for exactly that. Drop an image into the Style Reference slot in the prompt bar, or use –sref url on Discord, and Midjourney extracts the mood and visual treatment from it rather than its subject matter.
The strength of the effect is controlled by –sw, running from a low value that barely nudges the render toward the reference, up past the default toward a value where the style dominates almost everything else in the frame. Because the underlying algorithm behind style reference has been rebuilt more than once, older codes can render differently than they used to – –sv lets you pin which version of the algorithm a particular reference should use, which is worth knowing before assuming an old favourite code is broken rather than simply upgraded.
If you don't have a reference image at all, –sref random generates one for you, and it's genuinely random – running the same prompt with it twice in a row gives two unrelated styles. The value of doing this is less about the first result and more about what happens if you like it: the random style comes with its own code, and saving that code is the only way back to it. A style explorer page also lets you browse pre-rendered styles directly and search them by rough description – typing something like 'photographic' or 'anime' surfaces the codes that tend to match.
None of this replaces personalization – the two systems are meant to run together. A style reference sets the visual treatment; your personalization profile still shapes the finer choices the model makes inside that treatment, which is why the same style code can still look subtly different from one person's account to another's.
Chapter 8
Where the Community Comes In, and Where This Book Stops
A large amount of what Midjourney's models can do was shaped by the community rating images against each other rather than by an engineering team alone. Rating parties ask people to pick which of two images they find more beautiful, and that data trains both the default aesthetic and the personalization system that adapts it per person. Doing this occasionally is worth knowing about even if you never plan to participate – it's the reason the platform's baseline look shifts between versions the way it does.
Explore and the 'For You' feed use the same personalization signal to surface work other people have made that fits your own established taste, and a separate profile page lets you organise and show off your own images with a custom bio, banner and spotlighted pieces. Folders, introduced later, let you group ongoing projects rather than scrolling through one long, undivided feed of everything you've ever made.
A live collaborative 'Rooms' feature was tested and then removed from the website entirely, with Midjourney stating plainly that it tried to solve too many problems at once and wasn't built to scale – a useful reminder that features on this platform are genuinely experimental until stated otherwise, and can disappear as fast as they arrive.
A few limits are worth planning around before they surprise you mid-project. Upload size for reference and prompt images is capped, and an upload past that limit is simply rejected. Relax-speed rendering and HD video are both restricted to particular membership tiers, so a workflow built entirely around either one may not carry over cleanly if your plan changes. And because the underlying model is replaced every few months rather than every few years, a technique that works perfectly today is worth re-checking after the next version lands – the fastest way to do that is simply to read what changed, in Midjourney's own words, on its updates page.
Questions readers actually ask
Can I use a Style Reference and my Personalization profile in the same prompt?
Yes – the two are designed to run together. The style reference sets the overall visual treatment, and your personalization profile still shapes the finer choices the model makes inside that treatment.
Why does an old style reference code look different than it used to?
The underlying style-reference algorithm has been rebuilt more than once, and the newest version is the default. Add –sv followed by the version number to pin a code to the algorithm it was originally generated under.
If I use Omni-Reference with a heavily stylised prompt, why does the reference stop showing up?
Omni-Reference competes for influence with –stylize and –exp. A high value on either of those generally needs a correspondingly higher omni-weight (–ow) to keep the reference visible in the result.
Is a draft-mode image locked at low quality forever?
No. Clicking Vary on a draft re-renders it at full quality and resolution – the draft is only meant as a fast way to test which direction is worth finishing.
Can I animate an image I made somewhere other than Midjourney?
Yes. Drag the image into the prompt bar, mark it as the start frame, and describe the motion the same way you would for an image generated on the platform.
Does Character Reference work on a photo of a real person?
It's built around characters generated by Midjourney itself. Used on a real photograph it will distort the person in much the same way any ordinary image prompt does.
Contact / More useful information from RamthaMedia
Official source links:
Midjourney
As an Amazon Associate, RamthaMedia earns from qualifying purchases.
Disclaimer: This eBook is compiled from publicly available information and was accurate at the time of writing. For full and up-to-date details, please visit the official website linked above. RamthaMedia accepts no legal liability for any decision made on the basis of this eBook, and nothing here is professional, financial or legal advice. The image used for the cover page is illustrative only – a stock photo from Pexels or an AI-generated image, never a real photograph of the site described.
