FLUX.1 Tools Guide: Choosing Fill, Redux, Depth, and Canny

FLUX.1 Tools is Black Forest Labs’ collection of image-editing and conditioning models for FLUX.1. It includes Fill for masked editing, Redux for image variations, and Canny and Depth for structural guidance; downloadable variants are available through BFL’s Hugging Face repositories. BFL’s current announcement marks Canny and Depth as deprecated and unsupported in its API, while explicitly excluding Fill and Redux from that change. BFL announcement

The practical starting point is the input you already have. A mask, a reference photograph, and an edge map express different requirements, so choosing the appropriate tool matters before adjusting generation settings.

What are the key facts about FLUX.1 Tools?

Which FLUX.1 tool fits your image-editing task?

Fill accepts an existing image, a mask, and a description of the desired content. It supports editing within the image and extending an image beyond its original boundaries. This makes it a relevant starting point when the rest of a composition should provide context for a changed region. Fill model card

Redux starts from an image and produces variations. Its model card describes an adapter used with FLUX.1 base models, with a reference pipeline that passes image conditioning into a separate generator. That division matters: downloading the adapter alone does not provide a complete image-generation model. Redux model card

Canny and Depth encode structure differently. Canny uses detected edges; Depth uses a depth representation. BFL distributed full model and LoRA versions, and the relevant checkpoints remain available for reference despite the API deprecation. Announcement

As a workflow recommendation, write down the constraint before selecting a tool. “Replace this patch of background” suggests a mask. “Explore related versions of this composition” suggests image conditioning. “Retain these outlines” suggests structural guidance. These are selection heuristics based on the documented inputs, not guarantees of exact preservation.

What are the limitations of FLUX.1 Tools?

BFL’s deprecation notice affects the Canny and Depth families, including their LoRA versions. A downloadable checkpoint does not establish that the corresponding hosted endpoint remains supported. New API integrations should follow the current availability notice. BFL announcement

Fill can leave visible edge lines when replacing complex textures, and its model card notes possible color shifts outside the filled region. Inspect both the edited boundary and supposedly unchanged parts of the image. A narrow mask is not sufficient evidence that all surrounding pixels will remain identical. Fill model card

Redux outputs are strongly influenced by the input image, and the card warns that prompts may not be followed successfully. If your task requires a specific object change, evaluate whether reference-driven variation is the right operation before spending time tuning it. Redux model card

The Canny model is also not interchangeable with every component called ControlNet. Diffusers explains that BFL’s Canny Control uses a different conditioning architecture and a dedicated pipeline. Follow the example for the exact checkpoint, including whether it is the complete model or a LoRA. Diffusers documentation

How do you use FLUX.1 Fill and Redux locally?

Here is a local Fill example based on BFL’s model card. Install compatible PyTorch and the documented Diffusers dependencies, accept the repository’s access conditions, and authenticate your Hugging Face environment before downloading the gated weights. Fill card

Prepare source.png and a matching mask.png. In the documented Fill workflow, white identifies the area to regenerate and black preserves the surrounding context. Keep their dimensions aligned. Diffusers Fill pipeline

import torch
from diffusers import FluxFillPipeline
from diffusers.utils import load_image

source = load_image("source.png")
mask = load_image("mask.png")
pipe = FluxFillPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-Fill-dev",
    torch_dtype=torch.bfloat16,
).to("cuda")
edited = pipe(
    prompt="A blue ceramic vase on the wooden tabletop",
    image=source,
    mask_image=mask,
    width=source.width,
    height=source.height,
    guidance_scale=30,
    num_inference_steps=50,
).images[0]
edited.save("edited.png")

The guidance and step settings follow the card’s example; they are a starting point rather than universal optimal values. This CUDA example requires sufficient memory for the pipeline and your chosen image dimensions. It does not establish a minimum VRAM requirement. Fill card

For ComfyUI, use the official Fill workflow and its matching model files. Load your image, prepare the mask, select the documented loaders, and run the supplied graph before changing individual nodes. The tutorial separately explains inpainting and outpainting. ComfyUI Fill tutorial

For Redux, begin with the model card’s FluxPriorReduxPipeline plus FluxPipeline example. Keep that first run focused on image variation. Add more conditioning only after you understand which input is controlling the result. Redux card

Save the source, mask or reference, prompt, model identity, and output as a single experiment record. For a revision request, change one part of that record at a time so you can explain why the result changed. This is a suggested review practice, not a model feature.

How do Fill and Redux compare with Ultra?

Read the ComfyUI complete guide for working with saved graphs. The sibling FLUX1.1 Pro Ultra guide covers the hosted generation alternative.

What else should you know about FLUX.1 Tools?

Is FLUX.1 Tools one model download?

FLUX.1 Tools contains separate Fill, Canny, and Depth models plus the Redux adapter. Each tool has its own inputs and loading procedure. BFL announcement

Can I still use FLUX.1 Canny and Depth through BFL's API?

BFL marks FLUX.1 Canny and Depth as deprecated and unsupported in its API. Their open-weight checkpoints remain available for reference. BFL announcement

Does FLUX.1 Fill preserve everything outside the mask exactly?

The FLUX.1 Fill-dev model card warns that colors can shift outside the filled region. Compare the complete output with the original when preservation matters. Fill card

Can FLUX.1 Redux generate an image on its own?

FLUX.1 Redux-dev is an adapter that requires a FLUX.1 base model. Its official example loads both a Redux prior and a separate image-generation pipeline. Redux card

Sources

• BFL Tools announcement and deprecation notice
• FLUX.1 Fill-dev model card
• FLUX.1 Redux-dev model card
• Diffusers FLUX pipeline documentation
• ComfyUI Fill workflow
• FLUX1.1 Pro Ultra documentation

Related guides on PromptZone

• Realistic Photos with FLUX
• Best SDXL Models in 2026