然而，Midjourney 很可能使用與 Dall-E 2 和 Stable Diffusion 相似的系統，特別是因為兩家公司都詳細解釋了他們的訓練方法。 這些模型已經學會了圖像和用於描述它們的文本之間的共享關係。 Midjourney 過去曾表示，與其競爭對手一樣，它在互聯網上搜索圖像和文字來描述它們，並使用數十億張已發布的圖像進行訓練。
大多數具有公共源代碼的 AI 圖像生成器通過稱為擴散的過程生成這些圖像。 從本質上講，這些模型的工作原理是向圖像添加噪聲，將其減少為像素化的混亂。 然後它學會通過反轉噪聲過程來恢復這些數據。 這個過程一遍又一遍地重複，訓練模型添加噪聲，然後再次將其移除。 然後，該模型可以應用去噪過程，通過對圖像進行微小變化來創建逼真的圖像。 這與模型對圖像和用於描述它的文本之間關係的理解相結合，幫助 AI 圖像生成器通過提示理解您的要求。
Midjourney is an AI image generator that offers a unique artistic style different from its competitors like Dall-E 2. It generates dream-like, gothic-style paintings rather than realistic photos. The creators of Midjourney have not provided much information about its background or training, but it is likely that it uses a similar training method to other AI generators that use text to describe images. This involves scraping the internet for images and their associated text to train the model to create images based on prompts.
Most AI image generators use a process called diffusion to generate images. This involves adding noise to an image and then training the model to remove it, resulting in a realistic image with small variations. The model's understanding of the relationship between an image and its description is also used to generate accurate images based on prompts.
(The main source and structure of this document are from MJ Official website. This doc has incorporated the experience shared by some expert users, and also my annoted comments. This doc is for personal reference only and should not be re-distributed.)
(本文資料及結構主要來自 MJ 官網，並加入各方高手的意見及個人的使用感受或實例，本文只可以作私人參考之用，不得轉載。)
Midjourney Documentation 使用手册
Quick Start
1: Join the Discord
2: Find a Newbies Channel
3: Use the /imagine Command :
Using /imagine
Type /imagine prompt: or select the /imagine command from the slash commands pop-up.
Type a description of the image you want to create in the `prompt` field.
Click return to send your message.
4: Processes the Job:
5: Upscale or Create Variations:
U1 U2 U3 U4
U buttons upscale an image generating a larger version of the selected image and adding more details.
V1 V2 V3 V4
V buttons create slight variations of the selected grid image. Creating a variation generates a new image grid similar to the chosen image's overall style and composition.
🔄
The 🔄 (re-roll) reruns a Job. In this case it would rerun the original prompt producing a new grid of images.
6: Rate Images:
🪄 Make Variations 🔍 Light Upscale Redo 🔍 Beta Upscale Redo ↗️Web Make Variations: creates a variation of the upscaled image and generates a new grid of four options.
Beta/Light Upscale Redo: redo the upscale using a different upscaler model.
Web: Open the image in your gallery on Midjourney.com
😖 😒 🙂 😍 Click a smiley button to rate any upscaled image on the Midjourney website or in Discord. Each day, the top 1000 image raters receive an hour of free, Fast Mode time. Rank your work or the work of others. Rank based on your personal style and opinion. Love the look, effort, colors, concept, or theme? Please give it a 😍!
7: Save Your Image: Click on the image to open it to full size, and then right-click and choose Save image. On mobile, long tap the image and then tap the download icon in the top right corner. All images are immediately available to view on midjourney.com/showcase/recent/ Sign In with Discord to view.
8: Subscribe to a Plan
Trial users have roughly 25 free Jobs. Jobs do not presently expire but will also not renew. To make more images use the /subscribe command in any Bot Channel to generate a personal link to the midjourney account page. Do not share this personal link with others. For prices and more information go to Subscription Plans
Next Steps
Work in Your Direct Messages
Public channels move too fast? Use the /imagine command while direct messaging with the Midjourney Bot. Images made within your direct messages are still subject to content and moderation rules and are visible on the Midjourney website.
Learn More About Prompts
Learn the basics of writing prompts, or explore how descriptions of artistic mediums, locations, and time periods change an image.
Learn How to Combine Your Own Pictures
Learn how to upload and blend your own images together with the /blend command.
Get Answers
Need help or have questions? Try these commands:
/help
Shows helpful basic information and tips about the Midjourney Bot.
/ask
Get an answer to a question.
For more help, visit the #trial-Support channel on the Midjourney Discord.
Using Discord
Discord InterfaceYou can interact with the Midjourney Bot on Discord. The Midjourney server has channels for working collaboratively, technical and billing support, official announcements, offering feedback, and discussions. The community is supportive, encouraging, and eager to share their knowledge with users just starting out.
Midjourney Discord: https://discord.gg/midjourney
User's Guide
VersionMidjourney routinely releases new model versions to improve efficiency, coherency, and quality. The latest model is the default, but other models can be used using the
--version or --v parameter or by using the /settings command and selecting a model version.
Different models excel at different types of images.
--version accepts the values 1, 2, 3, 4 and 5.
--version can be abbreviated --v
Newest Model
The Midjourney V5 model is the newest and most advanced model, released on March 15th, 2023. To use this model, add the --v 5 parameter to the end of a prompt, or use the /settings command and select 5️⃣ MJ Version 5
This model has very high Coherency, excels at interpreting natural language prompts, is higher resolution, and supports advanced features like repeating patterns with --tile.
Current Model
The Midjourney V4 model is an entirely new codebase and brand-new AI architecture designed by Midjourney and trained on the new Midjourney AI supercluster. The latest Midjourney model has more knowledge of creatures, places, objects, and more. It's much better at getting small details right and can handle complex prompts with multiple characters or objects. The Version 4 model supports advanced functionality like image prompting and multi-prompts.
This model has very high Coherency and excels with Image Prompts.
Version 4 Styles 4a, 4b, and 4cMidjourney Model Version 4 has three slightly different "flavors" with slight tweaks to the stylistic tuning of the model. Experiment with these versions by adding --style 4a, --style 4b, or --style 4c to the end of a V4 prompt. brbr --v 4 --style 4c is the current default and does not need to be added to the end of a prompt.
Note on Style 4a and 4b --style 4a and --style 4b only support 1:1, 2:3, and 3:2 aspect ratios.
--style 4c support aspect ratios up to 1:2 or 2:1.
Previous ModelsYou can access earlier midjourney models by using the --version or --v parameter or by using the /settings command and selecting a model version. Different models excel at different types of images.
prompt example: /imagine prompt vibrant California poppies --v 1
--version 1 :default model: 02/22–04/22 very abstract and painterlylow coherency.
--version 2default model: 04/22–07/22 creative, colorful, and painterly low coherency.
--version 3default model: 07/22–11/22 highly creative compositions, moderate coherency.
--hd (high definition) early alternative model, busy detailed and abstract, low coherency.
Niji ModelThe niji model is a collaboration between Midjourney and Spellbrush tuned to produce anime and illustrative styles. The --niji model has vastly more knowledge of anime, anime styles, and anime aesthetics. It's excellent at dynamic and action shots and character-focused compositions in general.
prompt example: /imagine prompt vibrant California poppies --niji
Notes on the --niji model Niji does not support the --stylize parameter. Use the /settings command and select Style Med to reset to the default style setting for all --niji prompts.
Niji supports multi-prompts or image-prompts.
Test ModelsOccasionally new models are released temporarily for community testing and feedback. There are currently two available test models: --test and --testp, which can be combined with the --creativeparameter for more varied compositions. prompt example: /imagine prompt vibrant California poppies --testp --creative
Notes on current test models --test and --testpTest models only support --stylize values between 1250–5000.
Test models do not support multi-prompts or image-prompts.
Test models have a maximum aspect ratio of 3:2 or 2:3.
Test models only generate two initial grid images when the aspect ratio is 1:1.
Test models only generate one initial grid image when the aspect ratio is not 1:1.
Words near the front of the prompt may matter more than words near the back.
How to Switch ModelsUse the Version or Test Parameter
Add the following to the end of your prompt.
--v 1,
--v 2,
--v 3,
--v 4,
--v 4 --style 4a,
--v4 --style 4b,
--test,
--testp,
--test --creative,
--testp --creative or --niji
Use the Settings CommandType /settings and select your preferred version from the menu.
1️⃣ MJ Version 1
2️⃣ MJ Version 2
3️⃣ MJ Version 3
4️⃣ MJ Version 4
🌈 Niji Mode
🤖 MJ Test
📷 MJ Test Photo
UpscalersMidjourney starts by generating a grid of low-resolution image options for each Job. You can use a Midjourney upscaler on any grid image to increase the size and add additional details. There are multiple upscale models available for upscaling an image.
The U1 U2 U3 U4 buttons under each image grid are used to upscale the selected image. Midjourney Dimensions and Sizes
All sizes are for square 1:1 aspect ratios.
** Max Upscale is an older resource-intensive upscaler and is only available when users are in Fast Mode.
Midjourney Model 5
The newest Midjourney Model Version 5 produces high-resolution 1024 x1024 px image grids without needing an additional step to upscale each mage.
When using Midjourney Model Version 5, the U1 U2 U3 U4 buttons under each image grid will separate the selected image from the initial image grid.
Regular (Default) UpscalerThe latest default upscaler increases image size while smoothing or refining details.
Some small elements may change between the initial grid image and the finished upscale.
Detailed UpscalerThe Detail Upscaler, creates a 1024px x 1024px image and adds many additional fine details to the image.
Images that have been upscaled with the detailed upscaler can be upscaled again using the Upscale to Max button for a final resolution of 1664px x 1664px. Upscale to Max is only available when in Fast mode.
The Detailed upscaler is the default for Midjourney Model Versions V1, V2, V3, and hd.
Beta UpscalerThe Beta Upscaler creates a 2048px x 2048px image without adding many additional details. The Beta Uspcaler is useful for faces and smooth surfaces.
Use the --upbeta parameter to change the behavior of the U1 U2 U3 U4 upscale buttons to use the Beta Upscaler.
Anime UpscalerThe Anime Upscaler is the default upscaler for the --niji model. It upscales images to 1024px x 1024px and is optimized to work well with illustrated and anime styles. Use the --upanime parameter to change the behavior of the U1 U2 U3 U4 upscale buttons to use the Anime Upscaler.
RemasterRemaster is an additional option for previously upscaled images made with the V1, V2, or V3 Model Versions. Remaster will upscale an image again using the --test and --creative parameters blending the composition of the original image and the coherency of the newer --test model.
Remaster any previously upscaled job by clicking the 🆕 Remaster button found beneth the original upscale. To Remaster very old jobs, use the /show command to refresh that job in Discord.
Remaster Notes
Only works on Aspect Ratio up to 2:3 or 3:2
Inconsistent results with Multi Prompts
Does not work with Image Prompts.
How to Switch UpscalersUse the Settings Command
⏫ Regular Upscale ⬆️ Light Upscale 🔥 Beta Upscale
Type /settings and select your preferred upscaler from the menu.
Use an Upscaler Parameter Add --uplight, --upbeta or --upanime to the end of your prompt.
Use an Upscale Redo Button
⏫ Regular Upscale ⬆️ Light Upscale 🔥 Beta Upscale
After you upscale an image you will see a row of buttons underneath the image that let you redo the upscale using a different upscaler model.
Command ListYou can interact with the Midjourney Bot on Discord by typing a Command. Commands are used to create images, change default settings, monitor user info, and perform other helpful tasks.
Midjourney Commands can be used in any Bot Channel, on private Discord servers where the Midjourney Bot has been permitted to operate, or in a direct message with the Midjourney Bot.
Commands
/ask
Get an answer to a question.
/blend
Easily blend two images together.
/daily_theme
Toggle notification pings for the #daily-theme channel update
/docs
Use in the official Midjourney Discord server to quickly generate a link to topics covered in this user guide!
/faq
Use in the official Midjourney Discord server to quickly generate a link to popual prompt craft channel FAQs.
/fast
Switch to Fast mode.
/help
Shows helpful basic information and tips about the Midjourney Bot.
/imagine
Generate an image using a prompt
/info
View information about your account and any queued or running jobs.
/stealth
For Pro Plan Subscribers: switch to Stealth Mode
/public
For Pro Plan Subscribers: switch to Public Mode
/subscribe
Generate a personal link for a user's account page.
/settings
View and adjust the Midjourney Bot's settings
/prefer option
Create or manage a custom option.
/prefer option list
View your current custom options.
/prefer suffix
Specify a suffix to add to the end of every prompt.
/show
Use an images Job ID to regenerate the Job within Discord.
/relax
Switch to Relax mode.
/remix
Toggle Remix mode.
Deprecated
/private (replaced with '/stealth')
/pixels
/idea
Command ListUser Info
Use the /info command to see information about your current queued and running jobs, subscription type, renewal date, and more.
Subscription
The Subscription section shows which plan you are subscribed to and your next renewal date.
Job Mode
Shows whether you are currently in Fast or Relaxed Mode. Relax Mode is only available to Standard and Pro Plan subscribers.
Visibility Mode
Shows whether you are currently in Public or Stealth Mode. Stealth Mode is only available to Pro Plan subscribers.
Fast Time Remaining
Shows your remaining Fast GPU time for the month. Fast GPU time resets monthly and does not carry over.
Lifetime Usage
Shows your lifetime Midjourney stats. Images include all types of generations (initial image grids, upscales, variations, remix, etc).
Relaxed Usage
Shows your Relax Mode usage for the month. Heavy Relax Mode users will experience slightly slower queue times. Relaxed usage amounts reset monthly.
Queued Jobs
Lists all Jobs queued to run. A maximum of seven jobs can be queued at the same time.
Running Jobs
Lists all Jobs currently running. A maximum of three jobs can run at the same time.
Use the Info Command
Type /info in any Bot Channel or your Direct Messages. Only you will be able to see your Info popup.
Settings and PresetsThe Settings command provides toggle buttons for common options like model version, style value, quality value, and upscaler version. Settings also has toggle buttons for the /stealth and /public commands.
1️⃣ MJ Version 1 2️⃣ MJ Version 2 3️⃣ MJ Version 3 4️⃣ MJ Version 4 5️⃣ MJ Version 5 🌈 Niji Mode 🤖MJ Test 📷 MJ Test Photo
Sets the Model Versions used for jobs.
Model Version 5 is only available to users with a Midjourney subscription.
🔥 Half Quality 🔥 Base Quality 🔥 High Quality (2x cost)
Sets the Quality Parameter used for jobs.
Half Quality = --q .5, Base Quality = --q 1, High Quality = --q 2.
🖌️ Style Low 🖌️ Style Med 🖌️ Style High 🖌️ Style Very High
Sets the Stylize Parameter used for jobs.
Style Low = --s 50, Style Med = --s 100, Style High = --s 250, Style Very High = --s 750,
⏫ Regular Upscale ⬆️ Light Upscale 🔥 Beta Upscale Sets the Upscaler used for jobs.
🧍♂️Public 🕵️ Stealth Switch between Public and Stealth modes. Corresponds to the /public and /stealth commands.
🎛️ Remix Switch to Remix mode
🐇 Fast 🐢 Relax Switch between Fast and Relaxed modes. Corresponds to the /fast and /relax commands.
Settings Note
Parameters added to the end of a prompt will override selections made using /settings.
Custom PreferencesCreate custom options using prefer commands to add commonly used parameters to the end of prompts automatically.
/prefer auto_dm Completed Jobs are automatically sent to Direct Message
/prefer option Create or manage a custom option.
/prefer option list View your current custom options.
/prefer suffix specify a suffix to add to the end of every prompt.
Prefer Option
/prefer option set
/prefer option set / mine / --hd --ar 7:4 creates an option called "mine" that translates to --hd --ar 7:4.
Using /imagine prompt vibrant California poppies --mine, is interpreted as /imagine prompt vibrant California poppies --hd --ar 7:4.
Leave the "value" field empty to delete an option.
/prefer option list list all options created with prefer option set. Users can have up to 20 custom options.
Prefer Suffix/prefer suffix automatically appends the specified suffix after all prompts. Use the command without a value to reset.
Command example: /prefer suffix/ --uplight --video
Only Parameters can be used with /prefer suffix,
prefer suffix --no orange is accepted
prefer suffix orange::-1 is not accepted
Show JobYou can use the /show command with the unique Job ID to move a job to another server or channel, revive a lost job, or refresh an old job to make new variations, upscale, or use newer parameters and features.
/show only works on your own jobs.
Find a Job ID
Job IDs are unique identifiers used for each image generated by Midjourney. Job IDs look like this: 9333dcd0-681e-4840-a29c-801e502ae424 and can be found in the first part of all image filenames, in the URLs on the website, and in an image's filename.
On The Web
You can find the Job ID of any image in your member gallery by selecting ... > Copy... > Job ID.
From The URL
Job IDs are the last part of the URL when viewing an image in your midjourney gallery.
https://www.midjourney.com/app/users/381590592095911946/?jobId=9333dcd0-681e-4840-a29c-801e502ae424.
From The File Name
Job IDs are the last part of the file name when viewing an image you have downloaded from your gallery.
User_cat_cloud_spirit_9333dcd0-681e-4840-a29c-801e502ae424.png
Using a Discord Emoji Reaction
React with the envelope emoji ✉️ to send a completed Job to direct messages. The direct message will include the image's seed number and Job ID. The ✉️ reaction only works with your own Jobs.
How to Use Show
Use the /show
BlendThe /blend command allows you to upload 2–5 images quickly and then looks at the concepts and aesthetics of each image and merges them into a novel new image.
/blend is the same as using multiple image prompts with /imagine, but the interface is optimized for easy use on mobile devices.
/blend works with up to 5 images. To use more than 5 images in a prompt use image prompts with /imagine.
/blend does not work with text prompts. To use text and image prompts together, use image prompts and text with /imagine.
/blend Options
After typing the /blend command, you will be prompted to upload two photos. Drag and drop images from your hard drive or add images from your photo library when using a mobile device. To add more images, select the optional/options field and select image3, image4, or image5. The /blend command may take longer to start than other commands because your images must be uploaded before the Midjourney Bot can process your request.
Blended images have a default 1:1 aspect ratio, but you can use the optional dimensions field to select between a square aspect ratio (1:1), portrait aspect ration (2:3), or landscape aspect ratio (3:2).
Custom suffixes are added to the end of /blend prompts, like any other /imagine prompt. Aspect ratios specified as part of the /blend command override aspect ratios within a custom suffix.
Blending Tip
For the best results, upload images that are the same aspect ratio as your desired result.
How to Use /blend
Parameter ListParameters are options added to a prompt that change how an image generates. Parameters can change an image's Aspect Ratios, switch between Midjourney Model Versions, change which Upscaler is used, and lots more.
Parameters are always added to the end of a prompt. You can add multiple parameters to each prompt.
Using an Apple device?
Many apple devices automatically change double hyphens (--) to an em-dash (—). Midjourney accepts both!
Basic ParametersAspect Ratios
--aspect, or --ar Change the aspect ratio of a generation.
Chaos
--chaos
No
--no Negative prompting, --no plants would try to remove plants from the image.
Quality
--quality <.25, .5, 1, or 2>, or --q <.25, .5, 1, or 2> How much rendering quality time you want to spend. The default value is 1. Higher values cost more and lower values cost less.
Seed --seed
Stop
--stop
Style
--style <4a, 4b or 4c> Switch between versions of the Midjourney Model Version 4
Stylize
--stylize
Uplight
--uplight Use an alternative "light" upscaler when selecting the U buttons. The results are closer to the original grid image. The upscaled image is less detailed and smoother.
Upbeta
--upbeta Use an alternative beta upscaler when selecting the U buttons. The results are closer to the original grid image. The upscaled image has significantly fewer added details.
Default Values (Model Version 4) and Default Values (Model Version 5)
Model Version Parameters
Midjourney routinely releases new model versions to improve efficiency, coherency, and quality. Different models excel at different types of images.
Niji
--niji An alternative model focused on anime style images.
High Definition
--hd Use an early alternative Model that produces larger, less consistent images. This algorithm may be suitable for abstract and landscape images.
Test
--test Use the Midjourney special test model.
Testp
--testp Use the Midjourney special photography-focused test model.
Version
--version <1, 2, 3, 4, or 5> or --v <1, 2, 3, 4, or 5> Use a different version of the Midjourney algorithm. The current algorithm (V4) is the default setting.
Upscaler Parameters
Midjourney starts by generating a grid of low-resolution image options for each Job. You can use a Midjourney upscaler on any grid image to increase the size and add additional details. There are multiple upscale models available for upscaling an image.
Uplight
--uplight Use an alternative "light" upscaler when selecting the U buttons. The results are closer to the original grid image. The upscaled image is less detailed and smoother. Upbeta --upbeta Use an alternative beta upscaler when selecting the U buttons. The results are closer to the original grid image. The upscaled image has significantly fewer added details. Upanime Use an alternative upscaler trained to work with the when selecting the U buttons. This upscaler was specifically created to work with the --niji Midjourney Model.
.......
....to be continued 未完待續...
沒有留言:
張貼留言