Inpainting in Midjourney: A Deep Dive, Secret Commands & Tips for the Long Awaited Feature!

Theoretically Media
22 Aug 202313:58

TLDRIn this video, Tim dives into the newly released inpainting feature in Midjourney, a tool for restoring or altering parts of images. He explains how to use the feature, its limitations, and uncovers some secrets. Tim demonstrates the inpainting process by editing various images, including removing unwanted elements, changing facial expressions, and adding new objects. He also discusses the 'slider method' for when inpainting doesn't respond as expected. Tim highlights the feature's ability to recontextualize subjects while maintaining the original background, showcasing its power when combined with zooming. He also touches on the 'chaos' and 'weird' commands, which can produce unpredictable but sometimes interesting results. The video ends with a call for viewers to share their experiences and tips with the inpainting feature.

Takeaways

  • 🎨 **Inpainting Feature**: Midjourney has introduced a new feature called inpainting which allows users to restore or modify parts of images without additional input.
  • 🖌️ **Technique Usage**: Inpainting is used for highlighting an area and re-prompting within that area, making it easier to correct unwanted elements like extra limbs.
  • 🔍 **Region Selection**: The 'Very Region' button is used for inpainting, allowing users to select areas of the canvas for Midjourney to reimagine.
  • 📈 **Remix Mode**: To prompt specific changes, users should enable Remix mode, which can be found in the forward slash settings.
  • 📷 **Image Requirements**: Inpainting works best with upscaled images and covers regions of 20 to 50 percent of the image, although smaller areas can also yield good results.
  • 🔄 **Undo Option**: An 'Undo Last Step' feature is available for users to revert changes made during the inpainting process.
  • 🖥️ **Interface Notes**: The inpainting module is accessible and resizable within the Discord window, but it is locked to the Discord interface.
  • 📚 **PDF Availability**: All visual content from the screen is available as a free PDF on Gumroad, with donations appreciated.
  • 🔍 **Zooming and Inpainting**: Inpainting works well in combination with zooming out, but cannot be used after panning, which might be a temporary limitation.
  • 🧞‍♂️ **Powerful Tool**: The inpainting tool is powerful for changing ethnicities, emotions, and adding props to images while maintaining the original background.
  • ⚙️ **Commands and Techniques**: Commands like 'chaos', 'stylize', 'style', and 'weird' can be used within inpainting for different effects, although 'image waiting' is not functional.

Q & A

  • What does the term 'inpainting' refer to in the context of image and video editing?

    -Inpainting refers to the technique of restoring damaged or deteriorated parts of images or video. It has its roots in the restoration of artwork but is now used in digital editing to fill in missing or unwanted parts of an image or video.

  • What is the purpose of the 'Very Region' button in Midjourney's inpainting feature?

    -The 'Very Region' button, also referred to as the inpainting tool, allows users to select certain areas of their canvas and have Midjourney reimagine it without additional user input, based on contextual clues in the surrounding image.

  • How does the inpainting feature handle the addition of new elements to an image?

    -The inpainting feature can add new elements such as props or change aspects like emotions and ethnicities. It uses contextual clues from the existing image to integrate the new elements in a way that is consistent with the overall scene.

  • What is the 'Remix Mode' in Midjourney, and why is it important for specific inpainting prompts?

    -Remix Mode is a setting in Midjourney that, when turned on, allows for more specific inpainting prompts. It enables users to instruct Midjourney to include particular elements or changes within the selected area of the canvas.

  • What are some limitations of the inpainting feature in Midjourney?

    -Some limitations include finicky selection tools, occasional strange poses when the bounding box is placed around the entire subject, and instances where the feature may ignore the user's prompt or produce unexpected results.

  • How does the inpainting feature work with upscaled images?

    -The inpainting feature only works with upscaled images. It does not function after the panning process, which suggests that there might be a temporary glitch or limitation in the current implementation.

  • What is the slider method in Midjourney's inpainting feature, and how can it be used?

    -The slider method involves issuing a prompt with a repeated phrase using double colons (::) to split the prompt. This technique can help achieve desired results when the inpainting feature does not respond to a single prompt.

  • What are some commands that can be used with the inpainting feature in Midjourney?

    -Some commands that work with the inpainting feature include 'chaos', 'stylize', 'style', and 'weird'. These commands can be used to modify the behavior of the inpainting process, although 'weird' can produce unpredictable results.

  • How does the inpainting feature handle lighting and reflections when adding new elements?

    -The inpainting feature impressively handles lighting, ensuring that new elements match the overall lighting and composition of the frame. However, it may not always accurately reflect the light source's position or the reflection's details.

  • null

    -null

  • What is the recommended region size for the inpainting feature to work best?

    -Midjourney advises that inpainting works best when covering a region of 20 to 50 percent of the image. However, good results can still be achieved with smaller regions.

  • How does the inpainting feature in Midjourney compare to Photoshop's liquify feature?

    -The inpainting feature in Midjourney is considered to be at least on par with, if not better than, Photoshop's liquify feature, especially when it comes to making subtle changes like adding smiles to faces without looking off.

Outlines

00:00

🖼️ Introduction to Mid-Journey In-Painting

The video begins with an introduction to the new mid-journey in-painting feature, which is a highly anticipated tool for restoring or modifying parts of images or videos. The host explains that in-painting allows for the restoration of damaged areas or the addition of new elements to an image by 're-prompting' within a highlighted area. The feature is demonstrated through an example where extra fingers are removed from an image. The host also discusses the 'very region' button for in-painting and the importance of having 'remix mode' enabled for more control. The capabilities of in-painting are showcased, including its ability to work with upscaled images and its limitations when used in conjunction with panning.

05:01

🎨 Advanced In-Painting Techniques and Results

The host delves into more advanced in-painting techniques, demonstrating how to change the subject's pose, expression, and add props within an image. The video shows how lighting and composition are preserved even when significant changes are made to the subject. The host also addresses the finicky nature of the lasso tool and its differences from similar tools in Photoshop. The discussion includes examples of changing a character's ethnicity, emotion, and adding elements like ninjas to a scene. The host emphasizes the power of in-painting when combined with zooming in and out for recontextualizing a subject while maintaining the original background.

10:01

🛠️ Troubleshooting and Command Usage in In-Painting

The video concludes with troubleshooting tips for when in-painting does not produce the desired result. The host introduces the 'slider method' as a technique to overcome in-painting's limitations. This involves repeating parts of the prompt to guide the AI towards the desired outcome. The host also lists the commands that are effective within in-painting, such as 'chaos', 'stylize', 'style', and 'weird', while cautioning that 'weird' can lead to unpredictable results. The video ends with a demonstration of how in-painting can alter reflections and other background details, showcasing the feature's ability to generate complex and contextually aware changes to an image.

Mindmap

Keywords

Inpainting

Inpainting is a technique used to restore damaged or deteriorated parts of images or videos. In the context of the video, it refers to a feature in a software called 'mid-journey' that allows users to highlight an area of an image and prompt the software to re-imagine that area based on contextual clues. It's used to fix unwanted elements, such as extra limbs in a picture, and is a powerful tool for image editing without manual intervention.

Mid-journey

Mid-journey refers to a specific software or tool that is being discussed in the video. It seems to be an advanced image editing tool that has recently implemented the inpainting feature. The video dives into how to use this tool, its capabilities, and the fun that can be had with it, indicating that mid-journey is likely a creative or artistic application.

Remix Mode

Remix mode is a setting within the mid-journey tool that, when enabled, allows users to make specific changes to parts of an image using inpainting. It is an important aspect of the inpainting feature as it provides more control over what the software generates within the selected area.

Upscaled Images

Upscaled images are those that have been enlarged or 'scaled up' from their original size, often to improve resolution or detail. The video mentions that inpainting works best with upscaled images, which suggests that the feature is capable of handling high-resolution editing tasks.

Free Draw Tool

The free draw tool is a feature within the inpainting module of mid-journey that allows users to manually select areas of the image they wish to edit. It's described as being a bit finicky, indicating it may require some practice to use effectively.

Lasso Tool

The lasso tool is another selection tool used in the inpainting process. It allows for more precise and irregular selections compared to the free draw tool. The video demonstrates its use in changing specific parts of an image, such as adding a red balloon to a character's hand.

Chaos and Weird

Chaos and weird are commands or settings within the mid-journey tool that can be adjusted to affect the outcome of the inpainting. A chaos level of 100 or a weird level of 50, as mentioned in the video, would result in more unpredictable and creative transformations of the image.

Niji

Niji seems to be a term used in the context of the video to describe a certain type of image or feature within mid-journey. It's mentioned in relation to the inpainting feature, suggesting that it might be a style or format that the software can handle.

Slider Method

The slider method is a technique mentioned in the video for getting the inpainting feature to generate desired results when it initially fails to do so. It involves repeating part of the prompt with additional instructions, effectively splitting the prompt to guide the software more effectively.

Post-editing

Post-editing refers to the process of making further adjustments to an image after the initial inpainting has been done. The video suggests that users might need to stack the original image with the inpainted image in an image editor to clean up smudged areas or blend elements together.

Contextual Clues

Contextual clues are the visual elements surrounding the area being inpainted that the mid-journey software uses to determine what to generate in the selected area. The software relies on these clues to maintain consistency with the rest of the image.

Highlights

Inpainting feature has arrived in Midjourney, allowing restoration of damaged or deteriorated parts of images or videos.

Inpainting can be used to highlight an area and re-prompt within that area for image restoration.

The 'Very region' button is used for inpaintings, allowing Midjourney to reimagine selected areas without additional user input.

Remix mode must be turned on to access specific inpaintings, which can be done via forward slash settings.

Inpainting works best with upscaled images and can cover regions of 20 to 50% of the image.

The feature can be combined with zooming out for effective use, but cannot be used after panning.

Inpainting can change character details, such as turning a typical character into a woman with sad eyes.

The lasso tool in the inpaint module can be used for more precise area selection, but it may be finicky to use.

Inpainting can be used to add props, change ethnicities, and modify emotions in a subject.

The slider method can be used to issue commands that may not initially work, such as adding a white wolf to a scene.

Commands that work within inpaintings include 'chaos', 'stylize', 'style', and 'weird', but be cautious with 'weird' as it can produce unexpected results.

Inpainting can produce impressive results, such as changing a man's ethnicity or a character's emotion with high accuracy.

Inpainting can sometimes ignore user prompts, but persistence and experimentation can yield the desired outcome.

The feature can handle complex changes like adding ninjas to a cinematic action scene or changing a character's prop.

Inpainting can be used to correct common issues like removing extra limbs or fingers from generated images.

The tool can create a hybrid of two different styles when combining the work of two photographers.

Inpainting is a powerful tool that can completely recontextualize a subject while keeping the background mostly the same.

The feature can work with niji, as demonstrated by changing a character from holding a pistol to a katana.

Inpainting can sometimes produce errors, such as the return of double limbs, which users should be aware of.