Flux.2 [Klein] 4B: Image Edit

A smaller foundation model with exceptional quality-to-size ratio. Ideal for local deployment, fine-tuning on limited hardware, and efficient generation and editing workflows.

Download Workflow Try in Cloud API documentation

Input

Image to edit

Edit instruction

Megapixels

Advanced settings

CFG

Default: 5

Negative prompt

Seed

Steps

Default: 20

Run

Output

Example

Call this model from your code

The request below uses the same prepared graph and settings as the Playground. Each submission can start a paid run; do not automatically retry a submission whose outcome is unknown.

Get API key

What you need

Runs on: Comfy Cloud
Key: COMFY_API_KEY
Your files: Uploaded by the code from their urls

Needs a paid Cloud plan and credits.

Download API JSON

# Python 3.10+: pip install comfy-sdk==0.4.0
import os

from comfy_sdk import Comfy

api_key = os.environ.get("COMFY_API_KEY", "").strip()
if not api_key: raise RuntimeError("Set COMFY_API_KEY")
os.environ.setdefault("COMFY_BASE_URL", "https://cloud.comfy.org")
client = Comfy(api_key=api_key)
workflow = client.workflows.from_json({
    "9": {
        "class_type": "SaveImage",
        "inputs": {
            "filename_prefix": "Flux2-Klein-4b-base",
            "images": [
                "75:65",
                0
            ]
        },
        "_meta": {
            "title": "Save Image"
        }
    },
    "76": {
        "class_type": "LoadImage",
        "inputs": {
            "image": "UPLOADED_image_FILENAME"
        },
        "_meta": {
            "title": "Load Image"
        }
    },
    "75:61": {
        "class_type": "KSamplerSelect",
        "inputs": {
            "sampler_name": "euler"
        },
        "_meta": {
            "title": "KSamplerSelect"
        }
    },
    "75:62": {
        "class_type": "Flux2Scheduler",
        "inputs": {
            "height": [
                "75:81",
                1
            ],
            "steps": 20,
            "width": [
                "75:81",
                0
            ]
        },
        "_meta": {
            "title": "Flux2Scheduler"
        }
    },
    "75:63": {
        "class_type": "CFGGuider",
        "inputs": {
            "cfg": 5,
            "model": [
                "75:70",
                0
            ],
            "negative": [
                "75:79:76",
                0
            ],
            "positive": [
                "75:79:77",
                0
            ]
        },
        "_meta": {
            "title": "CFG Guider"
        }
    },
    "75:64": {
        "class_type": "SamplerCustomAdvanced",
        "inputs": {
            "guider": [
                "75:63",
                0
            ],
            "latent_image": [
                "75:66",
                0
            ],
            "noise": [
                "75:73",
                0
            ],
            "sampler": [
                "75:61",
                0
            ],
            "sigmas": [
                "75:62",
                0
            ]
        },
        "_meta": {
            "title": "SamplerCustomAdvanced"
        }
    },
    "75:65": {
        "class_type": "VAEDecode",
        "inputs": {
            "samples": [
                "75:64",
                0
            ],
            "vae": [
                "75:72",
                0
            ]
        },
        "_meta": {
            "title": "VAE Decode"
        }
    },
    "75:66": {
        "class_type": "EmptyFlux2LatentImage",
        "inputs": {
            "batch_size": 1,
            "height": [
                "75:81",
                1
            ],
            "width": [
                "75:81",
                0
            ]
        },
        "_meta": {
            "title": "Empty Flux 2 Latent"
        }
    },
    "75:67": {
        "class_type": "CLIPTextEncode",
        "inputs": {
            "clip": [
                "75:71",
                0
            ],
            "text": ""
        },
        "_meta": {
            "title": "CLIP Text Encode (Negative Prompt)"
        }
    },
    "75:70": {
        "class_type": "UNETLoader",
        "inputs": {
            "unet_name": "flux-2-klein-base-4b-fp8.safetensors",
            "weight_dtype": "default"
        },
        "_meta": {
            "title": "Load Diffusion Model"
        }
    },
    "75:71": {
        "class_type": "CLIPLoader",
        "inputs": {
            "clip_name": "qwen_3_4b.safetensors",
            "device": "default",
            "type": "flux2"
        },
        "_meta": {
            "title": "Load CLIP"
        }
    },
    "75:72": {
        "class_type": "VAELoader",
        "inputs": {
            "vae_name": "flux2-vae.safetensors"
        },
        "_meta": {
            "title": "Load VAE"
        }
    },
    "75:73": {
        "class_type": "RandomNoise",
        "inputs": {
            "noise_seed": 432262096973502
        },
        "_meta": {
            "title": "RandomNoise"
        }
    },
    "75:74": {
        "class_type": "CLIPTextEncode",
        "inputs": {
            "clip": [
                "75:71",
                0
            ],
            "text": "Change the room around the person to a bright, airy Scandinavian loft on a sunny winter morning: white-washed timber walls, tall frosted windows with snow on the sills, pale birch floorboards, a cream sheepskin rug, a slim black floor lamp and a large fiddle-leaf fig in a ceramic pot. Soft, cool daylight fills the space. Keep the person in the exact same position, scale and pose in the same leather armchair, with identical face, hair, expression and clothing. Maintain the same camera angle, framing and perspective. Only replace the environment, and let the new daylight fall naturally on the person and the chair."
        },
        "_meta": {
            "title": "CLIP Text Encode (Positive Prompt)"
        }
    },
    "75:79:76": {
        "class_type": "ReferenceLatent",
        "inputs": {
            "conditioning": [
                "75:67",
                0
            ],
            "latent": [
                "75:79:78",
                0
            ]
        },
        "_meta": {
            "title": "Set Reference Latent"
        }
    },
    "75:79:77": {
        "class_type": "ReferenceLatent",
        "inputs": {
            "conditioning": [
                "75:74",
                0
            ],
            "latent": [
                "75:79:78",
                0
            ]
        },
        "_meta": {
            "title": "Set Reference Latent"
        }
    },
    "75:79:78": {
        "class_type": "VAEEncode",
        "inputs": {
            "pixels": [
                "75:80",
                0
            ],
            "vae": [
                "75:72",
                0
            ]
        },
        "_meta": {
            "title": "VAE Encode"
        }
    },
    "75:80": {
        "class_type": "ImageScaleToTotalPixels",
        "inputs": {
            "image": [
                "76",
                0
            ],
            "megapixels": 1,
            "resolution_steps": 1,
            "upscale_method": "nearest-exact"
        },
        "_meta": {
            "title": "Scale Image to Total Pixels"
        }
    },
    "75:81": {
        "class_type": "GetImageSize",
        "inputs": {
            "image": [
                "75:80",
                0
            ]
        },
        "_meta": {
            "title": "Get Image Size"
        }
    }
})
workflow.set_input("76", "image", client.assets.from_url("https://comfy.org/catalog-media/829a1667-e532-47cf-be3a-2fa95ef5c423/e7251bd1-e9ff-4db5-9abd-0122e81ee771"))

# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)
for output in job.get_outputs("9"):
    output.to_file(output.name)

API documentation

Try an example

About this workflow

The 'Flux.2 [Klein] 4B: Image Edit' workflow is designed to provide users with a robust and efficient method for image editing using the Flux.2 Klein model. This workflow leverages the model's impressive quality-to-size ratio, making it particularly suitable for local deployment and fine-tuning on limited hardware resources.

The workflow integrates several key nodes, including LoadImage for importing images, SaveImage for exporting results, and custom nodes like 7b34ab90-36f9-45ba-a665-71d418f0df18 and 65c22b29-59aa-496b-89c6-55a603658670, which handle specific image processing tasks. Additionally, the MarkdownNote node is used to provide inline documentation and guidance within the workflow, enhancing user understanding and ease of use.

Technically, the workflow operates by first loading an image using the LoadImage node, which is then processed through a series of transformations and edits defined by the custom nodes. These nodes are configured to apply specific editing techniques, leveraging the capabilities of the Flux.2 Klein model to maintain high-quality outputs. The final image is saved using the SaveImage node, ensuring that the edited image is stored efficiently.

This workflow is particularly useful for users who need to perform high-quality image edits without the need for extensive computational resources, making it an excellent choice for both hobbyists and professionals looking to optimize their image editing processes.

Flux.2 [Klein] 4B: Image Edit

Input

Inside the workflow

Call this model from your code

What you need

Try an example

About this workflow

Questions & answers

Input

Inside the workflow

Call this model from your code

What you need

Try an example

About this workflow

Questions & answers

What makes the Flux.2 Klein model suitable for local deployment?

Can I use this workflow for batch image editing?

How do I adjust the image edits applied by the workflow?

Is there documentation available for understanding the custom nodes?