23 Sept 202308:41

TLDRThe video transcript describes the process of generating images with an IP adapter, focusing on creating fan art of various anime characters. The creator discusses the challenges of capturing the distinctive design elements of anime characters and shares their experience with different parameters and control weights. They test the adapter with characters from 'Goblin Slayer' and 'Kiki's Delivery Service', noting the impact of reference images and the model used. The creator also comments on the acquisition of Studio Ghibli by Nippon Television, speculating on the future distribution of Ghibli films. The summary highlights the potential of IP adapters for personal enjoyment and creative exploration in generating anime-style images.


  • 🎨 The speaker is experimenting with generating images using an IP adapter, which is a tool for modifying the style of generated images.
  • 🔍 They are adjusting parameters such as denoising, resolution, and control weight to achieve better results.
  • 🌟 The first character tested is the priestess from 'Goblin Slayer', and the speaker finds a control weight of 1 to be effective.
  • 👀 The character's design is noted to be slightly childish with round eyes, but the result is considered cute and satisfactory.
  • 📈 The use of 'Reference Only' is suggested to improve the generation of characteristic eyes and faces, which can be model-dependent.
  • 🚫 The speaker notes that when using multiple IP adapters, only the upper units seem to be reflected in the image generation.
  • 🌈 A close-up image of the priestess is used in the 'Reference Only' unit, leading to a significant improvement in the generated image.
  • 🧙‍♂️ The 'High Elf' from 'Goblin Slayer' is also tested, with similar findings regarding control weight and the effectiveness of close-up references.
  • 📚 The speaker mentions enjoying 'Goblin Slayer' on Netflix and purchasing the comics, indicating a personal interest in the source material.
  • 🧹 The original witch character 'Kiki-chan' from 'Kiki's Delivery Service' is attempted, but the unique drawing style of Studio Ghibli proves challenging to replicate.
  • 📰 News is shared about Nippon Television acquiring Studio Ghibli, which is seen as a solution to business succession and potentially impacting future distribution.
  • 🍂 The speaker expresses a personal fondness for autumn and ends the transcript on a note that reflects the changing seasons.

Q & A

  • What is the main focus of the video?

    -The main focus of the video is experimenting with generating images using IP adapters by adding various anime characters and adjusting parameters to create fan art.

  • What is the significance of the control weight in the IP adapter?

    -The control weight in the IP adapter determines the influence of the original character design on the generated image. A higher weight means the generated image will more closely resemble the original character.

  • What model is the speaker using for generating images?

    -The speaker is using a model called 'anime mix' from Any Roller, which is their favorite model for this purpose.

  • What does the speaker find challenging about generating images with the IP adapter?

    -The speaker finds it challenging to accurately reproduce the distinctive design parts of anime characters, such as characteristic eyes and faces, which can be quite dependent on the model used.

  • What is the role of the 'reference only' unit in the IP adapter?

    -The 'reference only' unit is used to provide additional input to the model, helping to improve the accuracy of the generated image by focusing on specific details like the character's eyes or face.

  • Why does the speaker mention that only the upper units are reflected when using an IP adapter in combination?

    -The speaker mentions this because when multiple IP adapters are used, only the images from the higher-ranking units are referenced in the generation process, which can affect the final output.

  • What is the speaker's opinion on the generated image of the priestess from Goblin Slayer?

    -The speaker is satisfied with the generated image of the priestess, finding it nice and cute, and they would leave the character as is.

  • What does the speaker suggest about the use of successive XYZ plots with different parameters?

    -The speaker suggests that when generating successive XYZ plots with different parameters, the prompts written with wildcards become fixed, which can result in similar images being generated.

  • How does the speaker feel about the acquisition of Studio Ghibli by Nippon Television?

    -The speaker seems intrigued by the news and wonders if it might lead to Ghibli films being available on streaming platforms like Hulu, but they also express uncertainty about the future distribution policy.

  • What is the speaker's approach to generating images of characters with distinctive hairstyles?

    -The speaker tries to find the best control weight for the IP adapter and uses close-up images to improve the generation process, acknowledging the difficulty in reproducing certain hairstyles with only the prompts.

  • What is the speaker's final verdict on using the IP adapter for generating anime character images?

    -The speaker finds the process interesting for personal enjoyment and believes that with the right reference image and model, one can easily make the generated image look like an anime character's face.

  • What does the speaker express about their love for the season of autumn?

    -The speaker expresses a fondness for autumn, stating that it is their most favorite season and that as the night progresses, it starts to feel like autumn.



🎨 Experimenting with IP Adapters for Anime Character Fan Art

The speaker discusses their recent interest in generating images using IP adapters and shares their process of experimenting with different parameters and characters to create fan art. They detail the technical aspects, such as denoising levels, resolution, strength, and control weight, and mention the use of Text 2 Image for verification. The speaker uses the priestess from 'Goblin Slayer' as a test subject, adjusting the control weight to achieve a satisfactory result. They also touch on the limitations when using an IP adapter in combination with other units and the importance of the reference image. The model used is 'anime mix' from Any Roller, and the speaker shares their personal satisfaction with the generated images, as well as their experience with generating successive XYZ plots and the discovery of prompt fixation.


📺 Challenges in Reproducing Ghibli's Art Style and Studio Ghibli News

The speaker attempts to generate an image of the High Elf from 'Goblin Slayer' and discusses the challenges in reproducing distinctive hairstyles and character designs. They explore different control weights and reference images to improve the generation process. The speaker also shares their personal enjoyment of 'Goblin Slayer' and its influence on their consumption of related media. The paragraph transitions into news about Nippon Television's acquisition of Studio Ghibli, providing details on the business decision and its implications for the studio's management and future anime production. The speaker speculates on the potential for Ghibli films to be distributed on platforms like Hulu and reflects on the impact of such a change. The paragraph concludes with the speaker's appreciation for the ability to generate anime character resemblances using IP adapters and their fondness for the autumn season.



💡IP Adapter

An IP Adapter is a tool used in the context of this video to modify and generate images with specific characteristics of anime characters. It allows the user to input certain parameters and control the generation process to create fan art or stylized images. In the video, the IP Adapter is used to generate images of characters from 'Goblin Slayer' and 'Kiki's Delivery Service,' adjusting the control weight to achieve the desired look.


Denoising is a process in image generation that aims to reduce the noise or unwanted elements in an image to produce a cleaner, more refined output. In the video, a '1.5 denoising' level is mentioned, which likely refers to the strength of the denoising algorithm applied to the generated images to improve their quality.

💡High-Resolution Fix

High-Resolution Fix refers to a setting or feature that ensures the generated images maintain a high level of detail and clarity. In the script, it is mentioned alongside a resolution range of '640-720,' indicating that the generated images are optimized for this specific resolution.

💡Control Weight

Control Weight is a parameter in the image generation process that determines the influence of the input factors, such as the IP Adapter, on the final output. The video discusses finding a 'good place' for the control weight, which means adjusting it to achieve a balance between the original character design and the desired stylized output.

💡Text 2 Image

Text 2 Image is a method of image generation where textual descriptions are used to create visual representations. The video uses this method to verify the generated images by comparing the textual descriptions of the characters with the resulting images.

💡Goblin Slayer

Goblin Slayer is an anime series from which characters are used as examples in the video. The video discusses generating images of the 'priestess' and 'High Elf' from this series, showcasing how the IP Adapter and other parameters can be used to create fan art of these characters.

💡Reference Only

Reference Only is a setting in the image generation process that uses a provided image as a reference for the style or characteristics, without directly incorporating it into the generated image. The video discusses using a close-up image of the priestess from 'Goblin Slayer' in this mode to improve the generation results.

💡Any Roller

Any Roller is mentioned as the user's favorite model for generating images in this context. It is likely a specific algorithm or tool used within the image generation software to produce anime-style images. The video does not provide further details about Any Roller, but it is implied to be a preferred choice for the user's creative process.

💡Anime Mix

Anime Mix refers to a model or style of image generation that combines elements of anime character design. The video mentions not using an 'illustration-like model' this time, suggesting that Anime Mix is a more stylized, less realistic approach to generating anime character images.

💡XYZ Plots

XYZ Plots are a method of visualizing data in three dimensions, often used in various fields for data analysis. In the context of the video, it seems to refer to a series of image generations with different parameters, where the 'wildcards' represent variables that are changed to produce a range of results.

💡Nippon Television

Nippon Television is a broadcasting company in Japan that is mentioned in the video as having acquired Studio Ghibli, the studio known for producing animated films by Hayao Miyazaki. The acquisition is discussed as a potential solution to business succession issues and the future distribution of Ghibli's films.


Hulu is a streaming service mentioned in the context of potential future distribution of Studio Ghibli's films. The video speculates that with Nippon Television's acquisition, there might be a possibility of Ghibli films being available on Hulu, which is associated with Nippon Television.


Using IP adapters to generate images with various anime characters and adjusting parameters to create fan art.

Applying 1.5 denoising with high-resolution fix on 640-720 and control weight of 0.45 for the IP adapter.

Storing elements extracted with the tagger in the prompt for image generation.

Challenge of generating images due to distinctive design parts of anime characters.

Verification of generated images using Text 2 Image method.

First character tested is the priestess from Goblin Slayer, noted for her cute anime version.

Finding a control weight of 1 effective for the priestess character generation.

Using Reference Only for generating illustrations with anime checkpoints to control characteristic eyes and faces.

Observation that only upper units' images are referenced when using IP adapters in combination.

Improvement in image generation by inserting a close-up image of the priestess as a reference.

Preference for the Any Roller model called anime mix over illustration-like models.

Challenges in using reference only for generating images, with mixed results.

High Elf from Goblin Slayer's distinctive hairstyle proving difficult to reproduce with only prompts.

Experimenting with different control weights for the High Elf character generation.

Personal enjoyment in watching Goblin Slayer on Netflix and purchasing related comics.

Next character to be generated is the original witch Kiki-chan with a completely different drawing style.

Difficulty in generating images with the new model due to the unique drawing style of the witch girl.

News about Nippon Television acquiring Studio Ghibli to address business succession.

Nippon Television's commitment to respecting Studio Ghibli's values and supporting its management.

Speculation about the potential for Ghibli films to be distributed on Hulu following the acquisition.

The effectiveness of IP adapters in generating anime character resemblances, depending on the reference image and model used.

Personal satisfaction with the results and intention to subscribe to the channel for similar content.

The presenter's love for autumn and how it sets the mood for the evening.