ComfyUI - Hands are finally FIXED! This solution works with all models!

Scott Detweiler
18 Jan 202412:16

TLDRIn this video, the creator discusses a method to improve the depiction of hands in images using AI, overcoming previous challenges. They introduce a sponsored Gigabyte laptop, which enhances their streaming capabilities. The process involves using a depth map preprocessor and control net to refine hand images, ensuring the AI focuses solely on the hands. The creator shares their mistakes and solutions, providing viewers with a detailed guide to achieve better results in their own projects.


  • ๐ŸŽฅ The speaker is addressing issues encountered during a previous live stream and explaining how to fix hands in images using AI.
  • ๐Ÿ’ป Gigabyte has sponsored the channel and provided a 17x laptop equipped with a 48-card, which has been used during live streams and video production.
  • ๐Ÿ–ผ๏ธ The process begins with using a basic graph and a simple prompt to generate an image, which may have hand-related issues.
  • ๐ŸŒŸ The use of the word 'hands' in the prompt is highlighted as a popular method to correct hand-related issues in images.
  • ๐ŸŽจ A custom node for the empty latent and a standard case sampler are used, with a fixed seed for consistency during exploration.
  • ๐Ÿค– The 'mesh grafer' node is introduced as a key component for resolving hand issues, utilizing a small model to predict hand shapes and create a depth map.
  • ๐Ÿ” The depth map generated helps the model understand the layout of the hands, with lighter areas being closer to the camera and darker areas further away.
  • ๐Ÿ”— The control net is used to refine the image, with the depth map being inputted to guide the corrections, particularly focusing on the hands.
  • ๐Ÿ› ๏ธ A mask is created to isolate the hands for correction, ensuring that only the hand region is processed and the rest of the image remains unchanged.
  • ๐Ÿ”„ The use of different masking techniques, such as 'tight B boxes', is suggested to improve results and address issues like extra fingers.
  • ๐Ÿš€ The final recommendation is to upscale the corrected image for further refinement and to resolve any remaining issues.

Q & A

  • What was the main topic of the video?

    -The main topic of the video was about fixing hands in images using AI, specifically focusing on improving the quality of the hands in a portrait of a woman in a summer dress and a flower garden.

  • What was the issue the speaker encountered in their previous live stream?

    -The speaker encountered a couple of issues in their previous live stream related to fixing hands in images, which they have since resolved.

  • How does the speaker describe the effectiveness of the method they are teaching?

    -The speaker describes the method as effective in fixing hands at about 90 percent of the images, and mentions that it works well with various models and is relatively simple and quick.

  • What is the role of the sponsor, Gigabyte, in the video content?

    -Gigabyte sponsored the channel and provided a 17x laptop equipped with a 48-card, which the speaker uses during live streams and for creating videos. The laptop allows for high-quality artwork creation on the go.

  • What was the initial prompt used to generate the image with the hands?

    -The initial prompt was 'portrait of a beautiful woman in a summer dress and a flower garden waving her hands and excited'.

  • What is the purpose of the 'mesh grafer' node in the process?

    -The 'mesh grafer' node is used to identify and correct the hands in the image. It uses a small model to determine the correct hand shape and outputs a depth map to guide the fixing process.

  • Why is the control net used in the process?

    -The control net is used to refine the image, particularly the hands, using the depth map generated by the 'mesh grafer' node. It helps to ensure that only the hands are redrawn, not the entire image.

  • What mistake did the speaker make during their live stream that they wanted to avoid in the video?

    -The speaker mistakenly used the same seed for both the case sampler and the control net, which resulted in issues with the final output, specifically making the hands look 'crunchy'.

  • How does the speaker address the issue of hands being the wrong size or having too many fingers?

    -The speaker suggests using bounding boxes to focus the correction on the hands and potentially adjusting the mask expansion settings to better fit the hands and fingers.

  • What additional step does the speaker recommend after fixing the hands?

    -The speaker recommends using an upscaler, such as the 'ultimate upscaler', to improve the overall quality of the image, particularly the face and to resolve any remaining issues.

  • How does the speaker plan to share the graph and other resources with the community?

    -The speaker plans to post the graph and other resources in the community area on YouTube for all the supporters of the channel, providing access to months of live streams and various other materials.



๐Ÿ–Œ๏ธ Fixing Hands in Artwork

The paragraph discusses a method to fix hands in images using a specific model. The speaker references a previous live stream where issues were encountered, but have since been resolved. The process is applicable to various models and is claimed to be simple and quick. The speaker also mentions a sponsored laptop from Gigabyte, used during live streams and videos, and praises its performance. A basic graph is introduced, and a specific model, the Juggernaut, is chosen for demonstration. The speaker uses a simple prompt to generate an image of a woman with incorrect hand depiction, aiming to correct it using a methodical approach rather than relying solely on the prompt. A custom node for the empty latent and a standard case sampler are used, with a fixed seed to control variables. The goal is to correct the hands without needing further intervention.


๐ŸŒŸ Addressing Common Errors

The speaker shares insights on common mistakes made during the process of fixing hands in images. They discuss a previous frustration with the process and aim to save the audience similar pain. The paragraph details the use of a model, the creation of a mask, and the importance of using a control net. The speaker emphasizes the need to mask the area of interest (the hands) to ensure only that part of the image is redrawn. They also discuss the limitations of the model, such as not knowing the correct hand size or fixing overly long fingers. A suggestion to use tight B boxes for the mask is provided to improve results.


๐Ÿ’ป Sponsored Content and Community Support

In the final paragraph, the speaker expresses gratitude towards Gigabyte for sponsoring the channel and providing a high-performance laptop. They also thank the community members for their support and provide an update on the resources available to them, such as live streams, graphs, and embeddings. The speaker encourages the audience to explore the community area on YouTube, where they can access these resources and support the channel further.



๐Ÿ’กfix hands

The process of correcting the depiction of hands in images, which is a common challenge in image generation models. In the video, the speaker aims to improve the accuracy of hand representation in images by using specific tools and techniques, making it a central theme of the content.

๐Ÿ’กGigabyte laptop

A brand of laptop computer mentioned as the sponsor of the video. The speaker uses this laptop for live streams and video creation, highlighting its capabilities and performance, especially with its 48-card setup, which signifies high processing power.

๐Ÿ’กJuggernaut model

A reference to a specific model or tool used within the image generation process. The Juggernaut model is one of the options that users can select to create their images, and it is implied that different models can be chosen based on personal preference.

๐Ÿ’กcontrol net

A term used in the context of image generation models, referring to a neural network that is trained to control the generation process. In the video, the control net is used to guide the model in creating accurate hand representations in the images.

๐Ÿ’กdepth map pre-processor

A tool or technique used to process images before they are input into a model, specifically to create a depth map that helps the model understand the layout of the hands in the image. This is crucial for the 'fix hands' process described in the video.

๐Ÿ’กcase sampler

A component of the image generation process that samples different variations of the image based on the model's predictions. In the context of the video, the case sampler is used to iterate through the image generation process, making adjustments to the hands with each step.


A value used in the image generation process to introduce randomness. The seed determines the starting point for the generation, and using the same seed allows for consistent results. In the video, the speaker emphasizes the importance of using different seeds for different iterations to avoid repeating the same mistakes.


A technique used in image editing to isolate specific parts of an image for modification while leaving the rest untouched. In the video, masking is used to focus the correction efforts on the hands, ensuring that only the hand area is processed and improved.


The process of increasing the resolution of an image, often to improve its quality or to fix issues that become more noticeable at higher resolutions. In the video, upscaling is suggested as a final step to refine the image further after fixing the hands.

๐Ÿ’กcommunity area

A platform or section where the video creator and viewers can interact, share resources, and support each other. In the context of the video, the community area is a space on YouTube where the channel's supporters can access exclusive content, files, and live streams.


Individuals or entities that provide financial support to the video creator, often in exchange for recognition or benefits. In the video, sponsors are acknowledged for their contributions and are granted access to exclusive content in the community area.


The speaker introduces a method to fix hands in images with a success rate of about 90%.

The process is applicable to various models and works well with 1.5 SDXL.

Gigabyte sponsors the channel and provides a 17x laptop used in live streams and videos.

The laptop features a 48-card setup, allowing for creation of impressive artwork on the go.

The speaker uses a basic graph and the Juggernaut model for the demonstration.

A simple prompt is used to generate an image of a woman with incorrect hand depiction.

The speaker emphasizes the importance of using a fixed seed for consistency in the demonstration.

The Mesh Grafter node is introduced as a key tool for identifying and correcting hand issues in images.

The use of a control net and depth map is detailed for refining the hand area in the image.

The speaker explains the creation of a mask to isolate the hand region for correction.

A case sampler is used to iterate and refine the image, focusing on the hand region.

The speaker discusses a mistake made in a previous live stream regarding the use of seeds.

The importance of using different seeds for the case sampler nodes is highlighted to avoid crunchiness in the corrected hands.

The speaker provides tips on adjusting mask settings to better handle hand size and finger length issues.

The use of bounding boxes to refine the mask around the hands is suggested for better results.

The speaker recommends upscaling the image after fixing the hands to improve overall quality.

The speaker expresses gratitude to Gigabyte for their sponsorship and support of the channel.

The speaker invites supporters to access exclusive content, including graphs and live streams from the community area on YouTube.