Decoding Midjourney GPU Time: A 7-Point Guide

Write by Nathan Clark

|

Update: 25/10/2023

• 7 Min Read Listen

Listen

A Reddit user once posed a question:

"......Anyway, I see stuff about fast GPU and relax GPU. It implies that a paid user cannot actually use it all they want, and they get a limited amount, like a cell phone data plan. Maybe I'm wrong? But even if I'm not, I don't understand at all how to evaluate just how much of something I get. What is an "Hour"? One would think an actual hour, but obviously, I'm not paying for only 3 hours a month. Next, I assume it's 3 hours of usage time, but what is usage time? Is the queue usage time?

What is the time it takes to make an image? On the other hand, maybe it's different per image and dynamic. But then, what's an average time?"

At the heart of Midjourney lies the GPU, driving its advanced image-generation prowess. One can unveil unparalleled efficiency and creativity by understanding its GPU time intricacies. This guide illuminates these details, empowering you to leverage the platform entirely. Dive in and elevate your Midjourney journey!

1. What Is GPU Time in Midjourney?

Midjourney is a premier AI-driven platform that converts textual prompts into captivating visuals. At its core, this prowess is facilitated by sophisticated graphics processing units (GPUs), which diligently process and translate these prompts into images.

Acquiring a Midjourney subscription provides users with a specific allocation of GPU processing time each month. "GPU time" doesn't pertain to the duration spent logged into Midjourney or Discord; instead, it quantifies the time Midjourney's GPUs expend on generating visuals based on user prompts. Typically, a standard image consumes approximately 1 minute of GPU time, with more intricate prompts demanding extended processing durations.

2. How GPU Time Is Consumed?

Key takeaways

• Fast vs. relax modes use GPU time differently.

• Most images take ~1 minute of GPU time.

• Factors that affect GPU time per image.

Midjourney offers two modes that consume GPU time differently:

Fast Mode

Fast mode provides priority access to Midjourney's GPUs. It aims to deliver your generated images as quickly as possible.

The amount of fast mode time you get depends on your subscription plan. For example, the Pro plan includes 30 hours of fast mode per month.

Most standard images take around 1 minute of fast GPU time. Complex images may take longer.

Relax Mode

Relax mode places your image generation requests in a queue. Delivery times are slower, but relax mode hours are unlimited for paying subscribers.

Relax mode is ideal when you're not in a hurry or to conserve your limited fast hours. Expected wait times are 1-10 minutes per image.

How Much Will Midjourney GPU Time per Image Cost?

In essence, the amount of GPU minutes utilized depends on multiple parameters ranging from the type of job to the quality setting chosen.

Typically, the Midjourney bot expends approximately one minute of GPU time to craft a single image. Nonetheless, specific conditions such as upscaling the image, opting for non-standard aspect ratios, or utilizing older versions of the Midjourney model can extend this duration. Conversely, generating multiple variations of an image or adjusting to lower-quality parameters can reduce the GPU time consumed.

Factors Influencing GPU Time 

The GPU time taken for a job can be categorized based on

• Job Type

• Aspect Ratio

• Quality Parameter

• Stop Parameter

Here are the details:

Understanding GPU Consumption Patterns

Gaining clarity on Midjourney's GPU usage nuances can assist in optimizing your tasks:

Job Types: Opting for variations, the standard square aspect ratio, or utilizing lower quality and stop parameters generally leads to reduced GPU time expenditure.

Aspect Ratios and Upscaling: Undertaking upscaling tasks or working with images with distinct aspect ratios, especially when using older model versions, can escalate GPU time consumption per image.

Frequent Commands: Commands like /imagine, when combined with standard quality and stop parameters, typically use up about a minute of GPU time for each image creation.

Given the rapid processing capabilities, the Midjourney bot can generate up to 60 images in an hour on fast GPU minutes. Furthermore, Midjourney offers the flexibility of producing unlimited images when users opt for the relaxed GPU time.

3. Monitoring GPU Time Usage in Midjourney

Be aware that GPU time expires. To prevent any loss, you need to monitor it regularly. Here are two methods to keep track:

1. Utilizing the /info Command

For a granular examination of GPU time for specific image generations:

1. Go to Discord, choose the Midjourney server, and then enter any of the rooms.

2. Input the /info command before initiating the process.

3. Once the generation is complete, type /info again.

This method allows you to assess the exact GPU minutes consumed by the task.

2. Accessing the Midjourney Account Page

For an overarching view of your GPU time:

1. Navigate to your Midjourney account via https://www.midjourney.com/account/.

2. Log in, and click “Manage Sub”.

    

3. Here, you can view the balance of your monthly fast hours.

Regular monitoring aids in ensuring you don't unexpectedly deplete your fast time allocation.

4. Midjourney GPU Times Plans

Midjourney provides a range of subscription options, each granting a distinct amount of GPU time:

5. What to Do When You Run Out of Fast Hours

Once your package's GPU time is depleted, Fast Mode image creation on Midjourney becomes unavailable. To continue leveraging Midjourney, consider the following alternatives:

Method 1 Switching to Relax Mode

After using up your fast hours, you can switch to Relax mode. You can continue generating images slower without any usage limit in this mode.

Only Standard, Pro, or Mega plan subscribers have access to this feature.

Method 2 Earning Free Hours from Rankings

Participate in Midjourney's Rank Pair challenges. A position within the top 2000 ranks will grant you bonus-free hours.

Method 3 Purchasing Additional Fast Hours

Should you find yourself short on time, additional fast hours are available for purchase at $4 per hour, consistent across all plans.

Method 4 Upgrading Your Subscription Plan

If you frequently exhaust your GPU minutes, consider transitioning to a more premium subscription tier.

6. Maximizing the Value of Your GPU Minutes

To optimize your GPU minutes and ensure you're using them judiciously:

1. Start with Lower Quality Settings

   Before finalizing an image, start with a lower-quality setting. This lets you preview the image without consuming as much GPU time. Once satisfied, you can then upscale and enhance the quality.

2. Experiment with Commands and Settings

   Midjourney offers a command suite allowing users to tailor the image generation process. Leveraging these commands can help in efficient GPU minute consumption.

   For instance, the Quality command plays a pivotal role in GPU consumption. By default, every render on Midjourney deducts from your GPU minutes. However, adjusting the quality using commands like /imagine --quality 0.25 can significantly reduce GPU time usage, speeding up the generation process by up to four times.

3. Utilize the /settings Command

   To fine-tune your image generation preferences, use the /settings command. This command presents a prompt with various options, allowing users to adjust style, quality, and upscaling settings. Certain options might be exclusive to specific subscription tiers.

4. Read the User Manual

   

   For a comprehensive grasp of Midjourney's capabilities, perusing the user manual is essential. Midjourney offers advanced commands that fine-tune a plethora of parameters.

   The manual elucidates these commands and intricate design options, guiding you to optimize settings for desired image outcomes. By understanding these nuances, you can maximize the efficiency and quality of your GPU time usage.

   Doc address:

   https://docs.midjourney.com/

7. FAQs 

1. What types of GPUs does Midjourney use?

   Midjourney uses Nvidia GPUs like the A100 and V100 via Google Cloud Platform servers. These provide the massive parallel processing power needed for AI image generation. 

2. How many images can I generate per GPU hour?

   You can expect to generate around 60 standard images per GPU hour in fast mode. But this can vary a lot based on image size, quality, and complexity. Using --q 0.25 may allow 240+ images per hour.

3. What should I do if I always run out of fast hours?

   If you consistently max out your monthly fast hours, consider upgrading to the next higher subscription tier. This gives you more included fast time each month. Buying extra hourly fast credits can also supplement a limited allotment.

4. Can I roll over unused fast hours to the next month?

   Unfortunately, unused fast hours do not roll over month to month. Your fast-hour allotment will be reset at the beginning of each billing cycle. So it's best to try to utilize them each month fully.

Conclusion

In closing, understanding GPU times is pivotal for maximizing the Midjourney experience. As we navigate this innovative realm of AI-driven image generation, being informed about GPU usage ensures efficiency and creativity. Whether you're a novice or a seasoned user, grasping these intricacies will undoubtedly enrich your Midjourney journey. Here's to crafting stunning visuals with informed choices!