『Reference Only』で“同じ顔”や”同じキャラクター”をキープできる。凄すぎるControlNetの新機能

Controlnet introduces a groundbreaking feature called 'Reference Only' that allows users to adjust various elements in an image while maintaining the same person or character. The process is straightforward: upload the original image, enable the feature, and select 'Reference Only'. The result is a series of images with similar faces but varying backgrounds and moods. Users can further refine the output by adjusting control modes and prompts. The feature is versatile, enabling the creation of cartoons, animations, and even frame-by-frame GIF movies. It also works well with different models, generating consistent character appearances across them. The 'Reference Only' feature is highly compatible with illustration systems, facilitating more detailed and accurate depictions. It can be used to change character attributes like hair color or to add props without losing the essence of the character. The technology also supports live-action images, offering a high-quality and flexible way to generate images with consistent faces but altered poses or clothing. Multi ControlNet is another feature that enhances the process by applying multiple controlnets for more complex adjustments. Overall, 'Reference Only' significantly lowers the barrier to creating the ideal image, opening up new possibilities for artists and designers.

Takeaways

• 🆕 Introducing 'Reference Only', a new feature in ControlNet that allows users to adjust different elements while maintaining the same character or face.
• 📸 To use 'Reference Only', you need to upload an image, enable the feature, and select it from the Preprocessor options.
• 🔍 The feature ensures that generated images closely resemble the reference image, even when the background or other elements change.
• 🎨 Users can adjust the priority of their prompt to generate images that are more similar to the reference, using Control Mode options.
• 🐕 When generating animals or characters, the 'Reference Only' feature can produce a series of images with consistent facial features.
• 🏃 Adding specific actions or poses to the prompt, like 'running', can be combined with 'Reference Only' to create dynamic character images.
• 🎭 The feature is versatile and can be used to create cartoons, animations, and frame-by-frame GIFs by increasing the batch count.
• 📈 The 'XYZ Plot' section from the Scripts offers another useful feature for applying prompts across different models, generating variations while keeping characters similar.
• 🖌️ 'Reference Only' is compatible with illustration systems, facilitating detailed depictions and making it easier to create cartoons and animations.
• 👩‍🦱 The feature allows for changes in character attributes like hair color, while keeping the character's face consistent across images.
• 🍎 Specific prompts can be adjusted for better accuracy, such as specifying the position of objects like an apple in a character's hand.
• 🎭 Multi ControlNet is a feature for applying multiple controlnets, which can be used to adjust poses and other elements in generated images.
• 🔧 Control Weight adjustments and Mode changes can help refine the generated images to better match the original character's face and atmosphere.
• 📈 Increasing the resolution and making other adjustments can significantly improve the quality of the generated images.

Q & A

What is the new feature introduced in ControlNet called?

The new feature introduced in ControlNet is called 'Reference Only'.

How does the Reference Only feature allow users to adjust different elements?

The Reference Only feature allows users to adjust different elements while keeping the same person or character in the generated images.

What are the steps to use the Reference Only feature?

To use the Reference Only feature, you start by selecting an image, then drop it into the ControlNet section, check the 'Enable' and 'Pixel Perfect' boxes, select 'Reference Only' from the Preprocessor, and finally press 'Generate'.

What happens if Reference Only is not visible in the Preprocessor options?

If Reference Only is not visible in the Preprocessor options, it indicates that ControlNet needs to be updated.

How can you increase the similarity of the generated image to the reference image?

You can increase the similarity by increasing the batch count, which allows you to view more generated images and select those that closely resemble the reference.

What is the effect of changing the Control Mode to 'My prompt is more important'?

Changing the Control Mode to 'My prompt is more important' gives higher priority to the prompt, resulting in a generated image that is very similar to the reference image.

How does ControlNet handle generating images with different poses or gestures?

ControlNet can generate images with different poses or gestures by adjusting the prompt to include the desired action, such as 'Running', while keeping the same facial features.

What other types of content can be created using ControlNet's Reference Only feature?

With Reference Only, users can create cartoons, animations, and even frame-by-frame GIF movies by increasing the number of batches.

How does the XYZ Plot section in the Scripts contribute to the image generation process?

The XYZ Plot section, while not detailed in the transcript, is implied to be a useful feature for applying prompts to different models, generating a variety of images while maintaining similarity across models.

What is the compatibility of the Reference Only feature with illustrative images?

The Reference Only feature is quite compatible with illustrative systems, facilitating more detailed depictions and aiding in the creation of cartoons and animations.

How can you change specific attributes of the character, like hair color, using Reference Only?

You can change specific attributes by adjusting the prompt to include the desired change, such as specifying 'blonde hair' or 'with red hair and an apple in your hand', and the feature will reflect these changes in the generated image while keeping the character's face.

What is Multi ControlNet and how does it enhance the image generation process?

Multi ControlNet is a feature that allows multiple controlnets to be applied, enhancing the flexibility of image adjustments. It can find a suitable image to match a specific requirement, such as holding an apple, and combine it with the original image to achieve the desired result.

Outlines

00:00

📷 Introduction to Reference Only Feature in Controlnet

The video introduces a new feature called 'Reference Only' in controlnet, which allows users to adjust various elements while maintaining the same character or person. The feature is considered one of the most impressive technologies within controlnet. The tutorial demonstrates how to use the feature by starting with an image of a dog and guiding through the process of enabling 'Reference Only' in the controlnet section. The result is a similar-looking dog image, showcasing the feature's ability to produce variations in mood and background. The video also explores different modes and emphasizes the feature's versatility in creating cartoons, animations, and GIF movies. It also touches on the compatibility of the feature with different models and its potential for detailed depiction in illustrations.

05:06

🎨 Advanced Usage of Reference Only and Multi ControlNet

The second paragraph delves into more advanced uses of the 'Reference Only' feature, including changing character behaviors and experimenting with different prompts and models. It demonstrates how to adjust the image to include new elements like running gestures or specific poses, such as sitting on a bench, while retaining the original face. The video also addresses the flexibility of the feature when it comes to changing clothing or hair color, and how to refine prompts for better accuracy. It introduces 'Multi ControlNet' for applying multiple controlnets and shows how to use Openpose settings to adjust poses. The video concludes by emphasizing the lowered barrier for generating ideal images and encourages viewers to explore the feature's many applications, inviting them to subscribe for more AI tool introductions.

Keywords

ControlNet

ControlNet is a technology or platform that allows users to manipulate and generate images based on specific prompts or references. In the video, it is highlighted as having a new feature called 'Reference Only' which enables the generation of images while maintaining the same character or face. This is a core concept as the entire video revolves around demonstrating the capabilities of this feature within ControlNet.

Reference Only

This is a new feature within ControlNet that allows users to keep the same person or character's face consistent across different generated images. It is significant because it provides a higher level of control over the generation process, enabling the creation of series or animations with a consistent character appearance. The video demonstrates how to use this feature by keeping the face of a dog or a woman consistent while changing other elements like the background or pose.

Pixel Perfect

The term 'Pixel Perfect' likely refers to the quality of the generated images, ensuring that they are detailed and accurate to the pixel level. In the context of the video, checking the 'Pixel Perfect' option would enhance the quality of the generated images, making sure that the details, especially the reference image's face, are maintained with high fidelity during the generation process.

Batch Count

Batch Count refers to the number of images generated in one go by the ControlNet system. Increasing the batch count allows for the generation of multiple images at once, which can be useful for creating variations or a sequence of images, such as frame-by-frame GIF movies. The video mentions increasing the batch count to demonstrate the consistency of the character's face across multiple images.

Control Mode

Control Mode is a setting within ControlNet that allows users to adjust the level of control they have over the image generation process. The video discusses changing the Control Mode to 'My prompt is more important' to give higher priority to the user's input, resulting in a more accurate representation of the reference image, particularly in generating a similar dog image.

Multi ControlNet

Multi ControlNet is a feature that enables the application of multiple ControlNets to a single image generation task. This feature is used to combine different prompts or instructions, such as holding an apple, to create a more complex and specific output. The video demonstrates how to use Multi ControlNet to adjust the pose of a generated image while retaining the original face.

Openpose

Openpose is a technology or setting within ControlNet that is used to apply or adjust the pose of a generated image. It is mentioned in the context of fixing the pose of an image while still retaining the original face and integrating new prompts. Openpose settings help in achieving a natural and consistent posture in the generated images.

Control Weight

Control Weight is a parameter that can be adjusted in ControlNet to balance the influence of different controls or references during the image generation process. In the video, the presenter lowers the Control Weight on the Openpose side to correct the face of the generated image, bringing it closer to the original reference.

Inpaint

Inpaint is a technique or tool used for editing images to fill in or change certain areas while keeping the rest of the image intact. The video suggests using inpaint for changing poses or clothes of a character in an image while maintaining the same face, offering more flexibility than traditional editing methods.

Illustrative Images

Illustrative Images refer to a type of visual content that is artistically rendered, often used in the context of cartoons or animations. The video discusses the compatibility of the Reference Only feature with illustrative systems, showing how it can be used to create detailed and consistent characters for cartoons and animations.

Live-Action Images

Live-Action Images are photographs or footage captured of real people or events, as opposed to animated or illustrated content. The video demonstrates the use of the Reference Only feature with live-action images, showing how it can generate images that closely resemble the face and characteristics of a live-action reference.

Highlights

Controlnet introduces a new feature called 'Reference Only' that allows users to adjust different elements while keeping the same person or character.

The 'Reference Only' feature is considered one of the most amazing technologies in controlnet.

To use 'Reference Only', users can select it from the Preprocessor after enabling the feature and checking the Pixel Perfect option.

Updating controlnet is necessary if 'Reference Only' is not visible in the Preprocessor options.

The resulting image with 'Reference Only' will have a similar appearance to the reference image.

Increasing the batch count can show variations in mood and background while maintaining similar faces.

Control Mode can be adjusted to prioritize the prompt, resulting in a more similar image generation.

Focusing on controlnet can yield good results if the right photo is selected.

The prompt can be modified to add gestures like 'running' while keeping the same face.

Controlnet's versatility allows for creating cartoons, animations, and frame-by-frame GIF movies.

Different models can be applied with the same prompts to generate a variety of images while keeping the subjects similar.

The 'Reference Only' feature is compatible with illustration systems, facilitating detailed depictions.

The hair color of a character can be changed while maintaining the character's identity.

Increasing the number of batches and careful selection can improve the accuracy of the generated image.

Specific instructions can be added to the prompt to generate images with particular attributes, such as holding an apple.

Multi ControlNet allows for the application of multiple controlnets to find a suitable image for a specific attribute, like holding an apple.

Openpose settings can be used to adjust poses in generated images.

Adjusting Control Weight and Mode can help retain the original face while changing other attributes.

The 'Reference Only' feature, combined with other functions, significantly lowers the barrier for generating an ideal image.

There are many more applications for 'Reference Only', encouraging users to explore and experiment.