After 10 months of silence, Midjourney V7 is finally released

After about 10 months of silence, Midjourney finally officially released its latest version, the V7 model. Midjourney V7 has a new and improved diffusion model architecture, which not only significantly improves the quality of image generation, but also achieves breakthrough progress in generation speed, output resolution, prompt word control, and style diversity.

As a leader in the field of AI painting, Midjourney V7 has once again refreshed the boundaries of AI painting with its higher image quality, faster generation speed and stronger personalization capabilities.

Midjourney V7 interface screenshot

01 Midjourney V7 Introduction

Midjourney V7 uses a new latent diffusion model that combines the advantages of the diffusion method and the generative transformer, resulting in significantly improved details and overall artistic expression of the generated images.

The advantage of this new diffusion model is that, through its latent diffusion model, the system can better capture image details and achieve stable generation, which not only maintains the original details of the image but also ensures the consistency of the generated style, greatly improving the user's creative experience.

V7 not only increases the generation speed by 20%-30%, but also has a qualitative leap in image details, texture and consistency. Especially when dealing with complex elements (such as human bodies, hands, objects), V7 performs particularly well, avoiding the "uncanny valley effect" common in previous versions.

Midjourney V7 supports a maximum output resolution of 4096×4096 pixels, providing users with higher-precision image generation capabilities. At the same time, the new version has about 50% more training data than the previous version, which provides greater support for the model to understand complex prompts and generate high-quality images. V7 also supports custom aspect ratio adjustment, which can generally be adjusted from 1:1 to 16:9, which can meet the needs of most image ratios.

In addition, V7 also introduced the "Draft Mode", which increases the generation speed by 10 times and reduces the cost by 50%, making it particularly suitable for rapid iteration and creative exploration.

02 Midjourney V7 Features

02.1 Added new style preset parameters

Midjourney V7 provides users with more options for generating styles. New style preset parameters include:

• --style random

• --style expressive

• --style cute

• --style raw

• --style 4k

By setting parameters, users can finely control the style and effect of the generated images.

02.2 Weight Control Techniques

In order to achieve fine control of screen elements, V7 introduces an improved weight control mechanism:

Use double colons (::) to control weights

Users can adjust the importance of each element by using the :: symbol followed by a specific value in the prompt word. By default, each prompt word has a weight of 1; for example, entering cat::2 sofa means that cat has a weight of 2, while sofa retains the default weight of 1.

• You can use negative numbers (such as cloud::-1) to reduce the influence of an element appropriately.

• The comma acts as a soft break, while :: is a hard break, allowing the weights of different phrases to be adjusted independently.

02.3 --testp parameter optimization

Compared with V6, the --testp parameter in V7 has been significantly improved in the details and style consistency of the generated images, and the generated images are more stable and delicate.

02.4 Seed parameter (seed)

02.4.1 Fixed seed to ensure consistency: Using a fixed --seed parameter can ensure consistency between multiple generation results, which is especially important when batch generating the same topic.

02.4.2 Recording and Reuse: When creating in batches, it is recommended to record the seed parameters used for each generation, and keep the seed unchanged when comparing the effects of different prompt words, so as to more accurately observe the impact of parameter changes.

Below is a table of common Midjourney parameters that help you have fine-grained control over composition, style, and quality of the resulting image:

03 V7 Draft Mode

Draft mode is a new highlight of V7, suitable for quickly generating creative prototypes.

• Speed ​​and Cost: The time and resource consumption is only 50% of the standard mode, and the image rendering speed is 10 times faster than the V6 version. Images can be generated faster and at a lower cost, which is suitable for rapid verification of ideas and creative iterations.

• Interactive interface: Once enabled, the web editor will automatically switch to a "conversational interactive interface", allowing users to edit image content in real time through simple commands. For example, by inputting commands such as "replace the cat with an owl" or "turn to night scene", the system can automatically adjust and generate new works based on the prompts.

• Voice function: supports voice recognition generation function. After clicking the microphone icon, users can shape pictures in real time through voice, and supports Chinese voice recognition, making creation more convenient and efficient.

• Advanced functions: Supports command extension, and can use advanced functions such as multi-version iteration and batch generation to provide users with more creative possibilities.

There are two ways to use the draft mode:

• 1. Click "Draft Mode" on the Midjourney website and use the voice input prompt.

• 2. Add the --draft parameter after the prompt, such as A futuristic city --draft.

However, the resolution of the images generated in this mode is lower than that of other modes, but its behavior and aesthetics are highly consistent with the standard mode. It is suitable for early concept development, brainstorming, and rapid generation of creative sketches. If the user is satisfied with the draft image, he can click the "Enhance" or "Change" button to re-render it with full quality.

In addition, the current super-resolution, editing, material replacement and other functions will still use the V6 model, and Midjourney will gradually upgrade these functions in the future.

Last but not least

From the new latent diffusion model, 4096×4096 high-resolution output, to the approximately 50% increase in training data volume and 30% increase in generation speed, every technical improvement in V7 provides greater support for creators to broaden their imagination and creative boundaries.

At the same time, through the addition of new style parameters, detailed weight control techniques and mature seed parameter practices, users can control the prompt word project more accurately and achieve a perfect balance between style and details.

However, Midjourney, you seem to be late. Now GPT-4o, Dream 3.0, Doubao and other AI drawing products have been very popular, and everyone’s expectations have been raised. Can you, the former king of the drawing field, make a comeback?