L2: Cool Text 2 Image Trick in ComfyUI - Comfy Academy

Olivio Sarikas
11 Jan 202413:07

TLDRThis tutorial from Comfy Academy guides viewers through creating an AI image workflow using a case sampler, which is the core of the process. The presenter demonstrates how to set up a model, inputs, and outputs to render AI images, including using a 'dream shaper 8' model and encoding prompts. The video also explores advanced techniques like using control nets for different light situations and creating variations of an image based on diverse prompts, showcasing the potential for creative and practical applications.


  • ๐Ÿ˜€ The video is a tutorial on creating a workflow for AI image generation using ComfyUI.
  • ๐Ÿ” Users can download or run the workflow in the cloud for free through the provided buttons on the Open Art platform.
  • ๐ŸŽจ The 'case sampler' is considered the core of the AI workflow and is the starting point for image rendering.
  • ๐Ÿ”Œ Inputs for the AI model include a checkpoint (AI model), positive and negative prompts, which are encoded for AI processing.
  • ๐Ÿ–ผ๏ธ The 'latent image' is an essential part of the workflow, representing the AI's internal data points that need to be decoded into pixels.
  • ๐Ÿ”‘ The 'vae decode' step is crucial for converting the latent image into a pixel-based image that can be viewed or saved.
  • ๐Ÿ–ผ๏ธโ€๐Ÿ’ป The output options include saving or previewing the image, with the distinction that previewing does not save the image to the drive.
  • ๐Ÿ› ๏ธ Customization of the workflow is possible through settings like steps, CFG scale, k sampler type, and noise level.
  • ๐Ÿ”„ The 'Q prompt' button initiates the rendering process, and the process can be monitored and controlled through the interface.
  • ๐ŸŒ„ The video demonstrates creating multiple images with different prompts and settings, such as a landscape at different times of day.
  • ๐Ÿค– Advanced techniques like 'control net' are introduced for creating images with the same scene but different lighting or characteristics.
  • ๐ŸŒ The potential applications of the workflow are vast, including creating diverse versions of images for marketing or other purposes.

Q & A

  • What is the main purpose of the video?

    -The main purpose of the video is to guide viewers through building a simple workflow for creating AI-generated images using a tool like ComfyUI, with a focus on using the case sampler as the core component of the AI workflow.

  • How can viewers access the workflow demonstrated in the video?

    -Viewers can access the workflow by visiting OpenArt, where they can download the workflow or use the 'Lounge workflow' green button to run it in the cloud for free.

  • What is the role of the case sampler in the AI workflow?

    -The case sampler is considered the heart of the AI workflow. It is used to generate images based on the inputs provided, such as the AI model, positive and negative prompts, and other settings.

  • What is a checkpoint in the context of the AI model used for image rendering?

    -A checkpoint is the AI model used to render the image. It is selected from a list of available models and is crucial for determining the style and outcome of the generated image.

  • What is the purpose of the positive and negative prompts in the workflow?

    -Positive prompts guide the AI towards the desired characteristics in the image, while negative prompts help to avoid undesired elements. They are encoded into a format that the AI can understand and use to generate the image.

  • What is a latent image and why is it used in the workflow?

    -A latent image refers to the latent data points used by the AI, not the final pixel image. It is an intermediate step that needs to be decoded into actual pixels, which is done through a process called VAE decoding.

  • How can the viewer customize the image rendering process?

    -The viewer can customize the image rendering process by adjusting various settings in the case sampler, such as the number of steps, CFG scale, k-sampling method, and noise level.

  • What is the difference between 'save image' and 'preview image' in the workflow?

    -'Save image' will both display and save the rendered image to the user's drive, while 'preview image' only shows the image without saving it.

  • How does the video demonstrate creating multiple images with different prompts?

    -The video shows how to copy and paste the workflow components to create multiple processes with different inputs, allowing for the generation of multiple images based on various prompts.

  • What is a control net and how is it used in the workflow?

    -A control net is a tool used to modify specific aspects of an image, such as lighting or depth, without changing the overall scene. It is used by preprocessing the image and then applying it to control the AI's rendering process.

  • How can the workflow be applied to create images for different ethnicities or scenarios?

    -By changing the prompts to describe different ethnicities or scenarios, the workflow can generate images that are similar in composition but vary in the specific characteristics, such as the ethnicity of a person or the lighting of a scene.



๐Ÿ–Œ๏ธ Building the AI Workflow

The script introduces the process of setting up a basic AI workflow for image rendering. It instructs the user to download and run the workflow from OpenArt, emphasizing the convenience of not needing to install anything. The focus is on the case sampler, which is the core of the workflow. The user is guided through connecting various components, including the AI model (checkpoint), positive and negative prompts, and the latent image settings. The script explains the need for encoding text prompts into a format that the AI can process and the importance of the VAE (Variational Autoencoder) for decoding the latent image into pixel data. The user is also shown how to set up the output for saving or previewing the rendered image.


๐Ÿ”„ Customizing and Running the AI Rendering Process

This paragraph delves into customizing the AI rendering process by setting parameters such as the number of steps, CFG scale, and the k sampler. It details how to initiate the rendering process using the Q prompt button and the subsequent steps that lead to the generation of the image. The script introduces additional options like batch count for rendering multiple images and the AO queue for continuous rendering until stopped. The user is also shown how to monitor the rendering process and how to stop it if necessary. The paragraph concludes with a creative tip on duplicating the workflow setup to create multiple variations of an image based on different prompts.


๐ŸŒ„ Exploring Advanced Techniques with Control Net

The final paragraph introduces advanced techniques using Control Net for creating images with different lighting conditions while maintaining the same scene details. It explains how to preprocess the initial rendered image to create a depth map, which is then used to influence the rendering process through the Control Net preprocessor and the Apply Control Net function. The script demonstrates how this technique can be used to create variations of the same scene under different lighting conditions, such as day, night, and sunset. Additionally, it showcases how Control Net can be utilized to render the same subject with different ethnic backgrounds, highlighting its potential applications in fields like marketing for diverse representation.




A workflow in the context of the video refers to a series of connected steps or processes that are followed to complete a task or project. In the video, the presenter is demonstrating how to build a workflow for creating AI-generated images. This workflow includes various components such as the case sampler, model checkpoint, and output settings, which are all integral to the AI image rendering process.

๐Ÿ’กCase Sampler

The case sampler is described as the 'heart of the whole AI workflow' in the video. It is a tool used to generate different versions of an image based on specific prompts and settings. The case sampler allows for the input of both positive and negative prompts, which guide the AI in creating the desired image while avoiding undesired elements.

๐Ÿ’กModel Checkpoint

A model checkpoint in AI refers to a specific state of the AI model that has been saved during the training process. In the script, the presenter selects a checkpoint named 'dream shaper 8' to use for rendering the AI image. This checkpoint represents a particular version of the AI's learning that will influence the style and outcome of the generated image.

๐Ÿ’กPositive Prompt

A positive prompt is a text input that describes the desired characteristics or elements that the user wants to see in the AI-generated image. In the video, the positive prompt 'mountain landscape digital painting Masterpiece' guides the AI to create an image that matches this description.

๐Ÿ’กNegative Prompt

A negative prompt is used to specify what the user does not want to appear in the AI-generated image. It helps to refine the image by excluding certain elements. For example, the negative prompt 'ugly and deformed' in the script is used to ensure that the AI avoids creating images with these undesirable qualities.

๐Ÿ’กLatent Image

A latent image in the context of AI refers to the underlying data representation of an image before it is decoded into pixel form. The script mentions setting up a latent image with specific resolution and batch size parameters, which are then used by the AI to generate the final visual output.

๐Ÿ’กVAE Decode

VAE stands for Variational Autoencoder, and 'VAE decode' refers to the process of converting the latent image data into an actual pixel image that can be viewed or saved. In the video, the presenter uses a 'vae decode' component to transform the AI's latent data points into a visual representation of the image.

๐Ÿ’กBatch Size

Batch size in AI image generation refers to the number of images that are processed in one go. The script specifies a batch size of one, meaning the AI will render one image at a time. This setting can be adjusted based on the user's needs for efficiency or output quantity.

๐Ÿ’กControl Net

Control net is a feature used to manipulate specific aspects of an AI-generated image, such as lighting or depth. In the video, the presenter uses a depth control net to create variations of an image with different lighting conditions while maintaining the same overall scene and details.

๐Ÿ’กQ Prompt

The 'Q prompt' button in the video script is an action that initiates the rendering process of the AI image. Once clicked, the AI goes through the steps necessary to generate the image based on the inputs provided in the workflow, such as the prompts and model settings.


Upscaling in the context of image generation refers to the process of increasing the resolution or detail of an image. The script mentions the possibility of upscaling an image and then stopping the process if the user is not satisfied with the result before further processing, such as additional rendering steps.


Introduction to building a simple workflow in ComfyUI for AI image rendering.

Downloading and running a workflow from OpenArt with the Lounge workflow button for cloud-based processing.

Explanation of the case sampler as the core component of the AI workflow.

Connecting the AI model checkpoint for rendering images.

Use of positive and negative prompts with the AI to guide image generation.

Encoding text prompts into a format usable by the AI through 'clip text and code'.

Connecting the model's CLIP input to the positive and negative prompts for processing.

Setting up the latent image with resolution and batch size parameters.

The necessity of decoding latent images into pixel images using 'vae decode'.

Choosing between using the model's integrated VAE or a separate VAE for decoding.

Selecting the output method for the generated image: saving or previewing.

Customizing the case sampler settings for steps, CFG scale, K sampler, and noise.

Using the Q prompt button to initiate the rendering process and observe progress.

Exploring extra options like batch count and continuous rendering with AO queue.

Utilizing view Q to monitor active and pending processes, and canceling if necessary.

Demonstrating the creation of multiple rendering processes with different inputs.

Creating a workflow for generating images with different prompts and lighting conditions.

Introduction to using control net for rendering the same scene with different light situations.

Using control net preprocessors like depth maps to influence AI image rendering.

Applying control net with depth to maintain image details while changing lighting.

Practical applications of rendering images with different ethnic variations for marketing.

The End screen with a call to action for likes and future content.