Flux 2 Klein 4B - Image Edit
ComfyUI in the Cloud
Deploy Flux 2 Klein 4B - Image Edit, from
$0.60 / hour
Flux.2 Klein 4B Image Edit ComfyUI Workflow
A ComfyUI image editing workflow powered by Flux.2 Klein 4B, featuring an FP8 diffusion model, Qwen 3 4B text conditioning, a dedicated Flux.2 image encoder and decoder, and prompt-guided image transformations.
Workflow Overview
This workflow utilizes the Flux.2 Klein 4B model for AI-powered image editing within ComfyUI. It combines an FP8 diffusion model, Qwen 3 4B text conditioning, and a Flux.2 image encoder and decoder to support image transformations based on natural-language instructions.
Unlike a traditional text-to-image workflow, this configuration accepts an existing image as an input. The supplied image provides visual context, while the text prompt describes the desired modifications or transformations.
The workflow uses the Flux.2 Klein 4B FP8 model for image generation and editing operations. The selected VAE provides the image encoding and decoding components required for processing image representations. The displayed configuration includes a noise seed of 1111443136920027 and an IMAGE output connection.
Generation Settings
Workflow Type
Image Edit
Diffusion Model
Klein 4B
Model Precision
FP8
Text Encoder
Qwen 3 4B
Image Input
External
Prompt
Empty
VAE
Flux.2
Output Type
IMAGE
The workflow exposes an image input rather than separate width and height controls. Image resolution and internal sampling parameters are not visible in the displayed subgraph.
Model Configuration
Image Editing Model (UNet)
FP8flux-2-klein-base-4b-fp8
Flux.2 Klein 4B serves as the primary diffusion model responsible for the image generation and editing process. The model uses conditioning information derived from the editing prompt and the supplied image to guide the generation process. The FP8 model configuration reduces model memory requirements compared with higher-precision model weights. This makes the configuration suitable for image editing workflows where memory efficiency is an important consideration.
Text Conditioning Model
Qwen 3 4Bqwen_3_4b
The Qwen 3 4B model provides text conditioning for the Flux.2 Klein image editing workflow. It processes natural-language instructions describing the desired changes to the input image. These instructions are converted into conditioning representations that guide the diffusion model during the editing process. This allows users to describe image modifications through text rather than manually configuring individual editing operations.
Image Encoder / Decoder (VAE)
Flux.2 VAEfull_encoder_small_decoder
The selected Flux.2 VAE provides the encoding and decoding components used during image editing. Image encoding converts pixel-space image information into latent representations suitable for processing by the generation model. After the editing process, the decoder converts the resulting latent representation back into a visible image. The selected model is identified as full_encoder_small_decoder, indicating a full encoder and small decoder configuration.
Image Input Configuration
The workflow includes an external IMAGE input connection for supplying the source image that will be edited. This input allows images from other ComfyUI nodes to be passed directly into the editing subgraph.
Source Image Input
The image input can receive an image from a Load Image node, an upstream image generation workflow, or another image processing node. The supplied image provides the visual information used as the starting point for the editing process.
Image editing workflows can use the source image to preserve important visual characteristics while applying changes described in the prompt. The degree to which the original image is preserved depends on the internal generation settings and editing instructions.
Image Editing Prompt Configuration
The workflow includes a dedicated text prompt field for describing the modifications to apply to the supplied image. The prompt is processed through the Qwen 3 4B text conditioning model and used to guide the Flux.2 Klein diffusion model.
Prompt Input
The prompt field is empty in the displayed workflow configuration. Users can enter natural-language instructions describing changes to the image's appearance, objects, environment, lighting, colors, or composition.
The prompt field allows the same workflow to be reused for different image editing tasks without changing the selected diffusion model, text encoder, or VAE.
Noise Seed Configuration
The noise seed controls the random initialization used during the image editing process. The displayed configuration uses the seed value 1111443136920027.
| Noise Seed | 1111443136920027 |
| Seed Control | Configurable |
| Randomization Control | Available |
A fixed seed can help reproduce editing results when the source image, prompt, model configuration, and other generation settings remain unchanged. Changing the seed modifies the random initialization and may produce a different edited image.
Additional Parameters
| Workflow Type | Image-to-Image Editing |
| Image Input | External IMAGE Connection |
| Prompt Input | Empty |
| Noise Seed | 1111443136920027 |
| Diffusion Model Precision | FP8 |
| Image Encoder / Decoder | full_encoder_small_decoder |
| Output Type | IMAGE |
| Sampling Configuration | Not exposed in the visible subgraph |
Surrounding Nodes and Workflow Integration
The Flux.2 Klein 4B Image Edit subgraph combines image input handling, text conditioning, model selection, and noise seed configuration into a reusable image editing component.
The workflow begins with an input image supplied through the IMAGE connection. In a typical ComfyUI setup, this image can originate from a Load Image node, another generation workflow, or an upstream image processing component.
The supplied image is prepared for processing using the selected image encoding components. The Qwen 3 4B text conditioning model processes the editing prompt and provides conditioning information to guide the Flux.2 Klein 4B diffusion model.
The diffusion model performs the image transformation process using the source image information and text-based instructions. The configured noise seed controls the random initialization used during generation.
After editing, the selected Flux.2 decoder converts the resulting latent representation into a visible image. The edited result is exposed through the IMAGE output connection.
The output can be connected to downstream ComfyUI nodes for image preview, saving, or additional processing. Exact internal sampling nodes and upstream or downstream connections are not visible in the displayed screenshot.
Workflow Highlights
- ✓ Flux.2 Klein 4B: Uses the Flux.2 Klein 4B diffusion model for prompt-guided image editing.
- ✓ FP8 Model Precision: Uses an FP8 model configuration to help reduce memory requirements compared with higher-precision model weights.
- ✓ Qwen 3 4B Conditioning: Processes natural-language editing instructions and provides conditioning for image generation.
- ✓ Image Input Support: Accepts existing images through an external IMAGE connection for editing operations.
- ✓ Prompt-Based Editing: Provides a dedicated text field for describing desired image modifications.
- ✓ Flux.2 Image Encoder / Decoder: Uses the full_encoder_small_decoder configuration for image representation processing and decoding.
- ✓ Seed-Based Editing: Supports configurable random initialization with the displayed seed set to 1111443136920027.
- ✓ IMAGE Output: Exposes the edited image for preview, saving, or further processing.
- ✓ Reusable Subgraph: Organizes image input, model configuration, prompt conditioning, and output handling within a single ComfyUI component.
Workflow Summary
This ComfyUI image editing workflow combines the Flux.2 Klein 4B FP8 diffusion model with Qwen 3 4B text conditioning and a dedicated Flux.2 image encoder and decoder. It accepts an existing image and uses natural-language instructions to guide the image editing process.
The workflow uses the full_encoder_small_decoder VAE configuration and includes a noise seed value of 1111443136920027. The prompt field is empty in the displayed configuration, and the edited image is exposed through an IMAGE output connection.
The subgraph organizes image input handling, text conditioning, diffusion model selection, and image decoding into a reusable component. It can be integrated with larger ComfyUI workflows for image editing, transformation, and additional image processing operations.