Introduction
Hello everyone. In this article, we will explain the basic mechanism for creating AI tools with "Tensor Art". We will introduce the particularly important meanings of "workflow" and "node", how to set them up, and procedures.
What is an AI tool?
AI Tools is a node-based tool for visually designing AI image generation. It consists of a processing flow (workflow) that combines nodes (like pieces of a puzzle) to generate an image.
Main features of workflow
Intuitive operability: simply place and connect nodes with drag and drop.
Flexible configuration: Fully customize models (checkpoints), LoRA, prompts, and more.
Real-time generation: You can start image generation immediately after setting.
It is important to understand first!
What is a node?
A node is a "small unit responsible for one process" in image generation. For example, there are nodes such as "Load Checkpoint" that loads an AI model and "CLIP Text Encode" that analyzes prompts.
Basic configuration of a node:
Input: Materials that start processing (e.g. prompt or model).
Output: Passes the results of processing to the next node.
In an analogy, nodes are like the "parts" of a pipeline. Connecting these together creates the overall flow.
What is a workflow?
A workflow is a series of image generation flows designed by connecting multiple nodes.
For example, create the following flow:
Load an AI model (Load Checkpoint)
Analyze the prompt (Prompt Encode)
Generate an image (Sampler)
Save the image (Save Node)
Visually constructing these flows enables image generation in Tensor Art.
Image Generation Workflow in Tensor Art: Basic Configuration and Steps
Below, we will explain the basic workflow and the role of each node in detail.
Overview of the Basic Workflow
The basic configuration for image generation in Tensor Art is as follows:
Load Checkpoint (AI model): Select the base generative model.
→ Node name: Load Checkpoint
Encode prompt (generation instruction): Specify the direction of image generation.
→ Node name: CLIP Text Encode
Apply LoRA model (optional): Add style and features.
→ Node name: Load LoRA
Image generation process: Generate image based on prompt and model.
→ Node name: KSampler
VAE decode: Adjust generated image to make it human-readable.
→ Node name: VAE Decode
Save image: Save generated image to file.
→ Node name: Save Image
Detailed explanation and setting method for each node
🌸⬇️Let's use a workflow using FLUX as an example. ⬇️🌸
1. Load Checkpoint
Role: Loads the model that is the basis of AI image generation.
Settings: ckpt_name: Specify the model name you want to use.
Example: FLUX-1-dev-fp8 (recommended checkpoint for TensorArt).
2. Load LoRA (Add style)
Role: Apply LoRA model that adds specific features and style.
Settings: lora_name: Enter the name of the LoRA model you want to use.
strength_model and strength_clip: Model influence (1.0 recommended).
3. CLIP Text Encode
Role: Converts the content of the image to be generated (prompt) into a format that AI can understand.
Settings:
Example prompt: "futuristic cityscape, neon lights, digital painting".
4. KSampler (Central process of image generation)
Role: Generates the actual image based on the prompt and model.
Settings:
steps: Accuracy of generation (approximately 20-30).
cfg: Applicability of the prompt (usually 1.0).
sampler_name: Sampling method (e.g. Euler).
5. VAE Decode
Role: Converts the generated latent image into the final image data.
Note: Select a VAE that corresponds to the checkpoint.
6. Save Image
Role: Saves the generated image as a file.
Settings:
filename_prefix: Specifies the beginning of the image name (e.g. "TensorArt_").
Example of actual workflow: Node connection
Below is an example of an actual node connection. Image generation is possible by reproducing this flow in the Tensor Art node editor.
Load Checkpoint → Load the AI model.
Add Load LoRA if necessary and apply styles.
Enter prompts into CLIP Text Encode and set the generation content.
Use FluxGuidance (adjust guidance scale) to fine-tune the influence of the prompt.
Generate an image with KSampler.
Adjust the image through VAE Decode.
Finally, save the image with Save Image.
Frequently Asked Questions
Q1. What is the difference between Checkpoint and LoRA?
Checkpoint: A model that is the basis of AI image generation. It determines the overall style.
LoRA: A module for adding specific additional styles and fine features.
Q2. Is VAE required?
Basically used in conjunction with Checkpoint. Without VAE, the color and resolution of the image may not be displayed properly.
Summary
Once you understand how nodes and workflows work, you can create your own images just the way you want them. Use this guide to get started with a simple setup! 😆👍
⬆️This is an image generated using the workflow introduced this time 😊
Next steps
Try your own prompts and settings.
Combine multiple LoRAs to pursue originality.
Experiment with high resolution and special styles.
By publishing the completed workflow, you can have many people use it as your AI tool 👍
Enjoy your creative adventure with Tensor Art!
Side note: Tips for beginners
Understand the basics of nodes: First, understand what each node does.
Start with a simple workflow: Try a workflow with a minimum number of nodes to help you understand how it works.
Repeat the experiment: Adjust the parameters of each node and see how the generated image changes.
