Qwen Image 2.1 - Text to Image
ComfyUI in the Cloud
Deploy Qwen Image 2.1 - Text to Image, from
$0.60 / hour
Qwen-Image-2.1 ComfyUI Workflow
A ComfyUI image generation workflow powered by Qwen-Image-2.1, featuring INT8-quantized models, Qwen3-VL conditioning, and efficient sampling settings.
Workflow Overview
This workflow utilizes the Qwen-Image-2.1 model for AI image generation within ComfyUI. It combines Qwen3-VL 8B conditioning, a dedicated Qwen-Image-2.1 VAE, and an additional PE model designed for text-to-image generation.
INT8 quantization helps reduce memory requirements for selected model components. The workflow is configured with the Euler sampler, 25 sampling steps, a simple scheduler, and a CFG scale of 1.0.
Generation Settings
CFG Scale
1.0
Sampling Steps
25
Sampler
Euler
Scheduler
Simple
Width
External Input
Height
External Input
Cache Device
Auto
Data Type
Default
Model Configuration
Image Generation Model (UNet)
INT8qwen_image_2.1_int8_convrot
The primary image generation model responsible for the denoising process. This INT8-quantized version helps reduce memory consumption compared with higher-precision model weights.
Text / Vision Conditioning Model
Qwen3-VL 8Bqwen3vl_8b_int8_convrot
The Qwen3-VL 8B model provides conditioning information for the image generation process, helping interpret input instructions and visual context. It supports the generation model by converting input information into representations that guide the image generation process.
Variational Autoencoder (VAE)
BF16qwen_image_2.1_vae_bf16
The VAE converts generated latent representations into visible images. This workflow uses the BF16 Qwen-Image-2.1 VAE. The decoding stage transforms the latent output produced during sampling into an image that can be previewed, saved, or processed further.
Additional PE Model
INT8 / T2Iqwen3.5_9b_qwen_image_2.1_pe_t2i.int8_convrot
An additional INT8-quantized PE model is included in the configuration. Its identifier indicates a text-to-image component. Its exact function depends on the connected ComfyUI nodes and the internal workflow structure.
Additional Parameters
| Seed | 447606998181262 |
| Data Type | Default |
| Cache Device | Auto |
| Image Dimensions | Controlled through external width and height inputs |
Surrounding Nodes and Workflow Integration
The Qwen-Image-2.1 subgraph combines the model configuration and sampling parameters into a reusable generation component. The width and height parameters are exposed as input connections, allowing other nodes to control the dimensions of generated images.
In a typical ComfyUI workflow, prompt input and conditioning nodes provide instructions for image generation. The diffusion model performs the denoising process using the configured sampler and scheduler. The VAE then decodes the resulting latent representation into an image that can be connected to preview, saving, or additional processing nodes.
The subgraph structure helps organize multiple model components and generation settings while allowing integration with larger image generation workflows. Exact upstream and downstream connections depend on the surrounding ComfyUI node graph.
Workflow Highlights
- ✓ Qwen-Image-2.1: Uses the Qwen image generation architecture.
- ✓ Memory Efficiency: INT8-quantized components can help reduce model memory requirements.
- ✓ Vision-Language Conditioning: Integrates Qwen3-VL 8B for conditioning.
- ✓ Euler Sampling: Uses 25 sampling steps with a simple scheduler.
- ✓ Additional T2I Component: Includes an INT8 PE model with a text-to-image designation.
- ✓ Flexible Image Dimensions: Width and height are configurable through external node connections.
- ✓ Reusable Subgraph: Consolidates model components and sampling settings for integration into larger ComfyUI workflows.
Workflow Summary
This ComfyUI workflow configuration combines Qwen-Image-2.1 with Qwen3-VL 8B conditioning, INT8-quantized model components, and a BF16 VAE. It uses 25 Euler sampling steps, a simple scheduler, and a CFG scale of 1.0.
The workflow also includes an additional text-to-image PE model and externally connected image dimension inputs. Its subgraph design allows model configuration, sampling parameters, and image generation components to be organized within a reusable ComfyUI workflow.