To run FLUX.2 dev in ComfyUI, load the official text-to-image template and install the files named in its model notes. The inspected template combines Black Forest Labs’ 32B dev image transformer with a text encoder, a small-decoder VAE, and a Turbo LoRA. Model card, Text-to-image workflow
Start with text-to-image generation to check the installation before adding reference images.
What are the key facts about FLUX.2 dev in ComfyUI?
| Field | Verified detail |
|---|---|
| Developer | Black Forest Labs; ComfyUI publishes the integration tutorial. Release, ComfyUI guide |
| Released | November 25, 2025. Release |
| Type | Rectified-flow image transformer for generation and reference editing. Model card |
| Size or parameters | 32 billion for the dev image transformer. Text encoder and VAE are separate components. Model card, ComfyUI guide |
| License and access | Downloadable dev weights under BFL’s non-commercial model license; commercial licensing is separate. Release, Repository |
| Where it runs | Local ComfyUI using the documented model files; BFL also lists hosted dev providers. ComfyUI guide, Release |
Which FLUX.2 dev workflows does ComfyUI provide?
FLUX.2 dev handles generation, single-reference editing, and multiple-reference editing in one model. The official repository records these capabilities explicitly. Repository
ComfyUI’s tutorial offers separate examples for text-to-image generation, image editing, and product mockups. Pick the example that matches the inputs you have rather than beginning with the most elaborate graph. ComfyUI guide
A text-only first run gives you a simple installation check. After it succeeds, save the workflow and its settings before introducing another input or changing the component set.
The official dev card describes character, object, and style references without additional training. That makes reference editing an option to evaluate before preparing a custom training dataset. Model card
For a product mockup, write down which reference supplies the product and which supplies the design. Compare the result with both inputs, paying attention to geometry and placement.
ComfyUI’s graph interface also supports saving and loading workflows. Preserve a working graph as an installation artifact that you can return to when later experiments fail. ComfyUI repository
What should you check before running FLUX.2 dev in ComfyUI?
The component set is substantial. BFL’s Diffusers guide distinguishes quantized consumer-GPU configurations from larger-memory setups and describes CPU offloading and remote text encoding. Deployment guide
Those are specific implementations. Do not transfer a memory figure from a quantized Diffusers example to a ComfyUI graph using different components and assume it is a guaranteed minimum.
ComfyUI’s tutorial explicitly uses quantized weights. Keep the precision and filenames in your setup record so that later comparisons identify the same configuration. ComfyUI guide
The official ComfyUI page calls dev “open-source,” while BFL’s repository specifies a non-commercial model license. Use BFL’s weight-license description when deciding how to deploy this checkpoint. ComfyUI guide, Repository
Missing nodes are another installation constraint. The tutorial identifies outdated ComfyUI versions and failed node imports as possible causes; availability can differ between development and stable releases. ComfyUI guide
Use the startup log to identify the actual failure. A graph that cannot load a node needs a different investigation from a graph that loads successfully but runs out of memory.
How do you install and run the FLUX.2 dev ComfyUI template?
Update the ComfyUI installation you intend to use. Follow the installation-specific update instructions linked from the official tutorial. ComfyUI guide
Open that tutorial and choose its FLUX.2 Dev Text to Image workflow. It is documented as requiring no input image. ComfyUI guide
Download the diffusion model, text encoder, VAE, and Turbo LoRA linked in the downloaded JSON’s model notes. Place them in the directories shown below. Text-to-image workflow
ComfyUI/models/diffusion_models/flux2_dev_fp8mixed.safetensors
ComfyUI/models/text_encoders/mistral_3_small_flux2_bf16.safetensors
ComfyUI/models/vae/full_encoder_small_decoder.safetensors
ComfyUI/models/loras/Flux_2-Turbo-LoRA_comfyui.safetensors
- Load the workflow and check that its model controls select those files. Keep the template’s graph and sampling settings together for the first run. Text-to-image workflow
The tutorial’s general file list names flux2-vae.safetensors, while the inspected text-to-image JSON selects full_encoder_small_decoder.safetensors and includes an active Turbo LoRA loader. Follow the JSON’s model notes for this setup. ComfyUI guide, Text-to-image workflow
This is an adapted dev configuration. For fal’s separate eight-step adapter recipe, see the FLUX.2 Turbo and Flash guide. Turbo adapter card
- Enter a short scene description and run the graph. Save the resulting image and workflow together after a successful execution. ComfyUI guide, ComfyUI repository
For a manual checkout whose dependencies are already installed, launch ComfyUI from its directory with the documented command: ComfyUI repository
python main.py
Desktop and portable installations have their own launch paths. Use the method documented for your package instead of treating this command as a universal installer. ComfyUI repository
An original first prompt could be: “A yellow ceramic bowl on a dark stone counter, window light from the left, simple kitchen background.” Choose an image whose objects you can inspect easily.
If a file is absent from a selector, check the download location, exact filename, and component category. Compare those details to the official list before downloading unrelated checkpoints.
If the graph reports a missing node, check the supported ComfyUI version and read the startup import errors. These are the failure categories highlighted by the tutorial. ComfyUI guide
If it runs out of memory, record the graph, model files, image dimensions, hardware, and error. Compare complete documented configurations before making a hardware purchasing decision.
For the broader download decision, see the sibling FLUX.2 dev access guide. Keep a copy of the working baseline while evaluating alternatives.
Once text generation works, open the tutorial’s image-editing or product-mockup example. Supply the images requested by that template and preserve its original reference connections for the first test. ComfyUI guide
For your own edit, choose one visible alteration and record what should stay unchanged. Inspect the whole output, including the background, instead of looking only at the edited object.
How does ComfyUI compare with other FLUX.2 setup options?
| Route | Main setup distinction |
|---|---|
| FLUX.2 dev in ComfyUI | The inspected text-to-image graph loads diffusion, encoder, VAE, and Turbo LoRA files. Text-to-image workflow |
| FLUX.2 dev in Diffusers | A Python pipeline with documented quantization and offloading configurations. Deployment guide |
| Klein in ComfyUI | A different BFL family with smaller image transformers and separate templates. Klein guide |
Read the sibling Klein editing walkthrough to evaluate that option. The ComfyUI pillar provides interface background.
What else should you know about FLUX.2 dev in ComfyUI?
Is the FLUX.2 dev checkpoint enough for the ComfyUI template?
The inspected FLUX.2 dev text-to-image template also loads a Mistral text encoder, a small-decoder VAE, and a Turbo LoRA. Install all four files named in that template’s model notes. Text-to-image workflow
Which FLUX.2 dev workflow needs no reference image?
ComfyUI’s FLUX.2 Dev Text to Image example requires no input image. Select that template when starting from a text prompt. ComfyUI guide
Why are nodes missing in a FLUX.2 dev template?
ComfyUI’s FLUX.2 dev guide identifies outdated builds and startup import failures as possible causes of missing nodes. Check the startup log and the version required by the template. ComfyUI guide
Does the ComfyUI template run FLUX.2 pro?
The documented ComfyUI template loads FLUX.2 dev components and a Turbo adapter. FLUX.2 pro is a separate hosted offering in BFL’s model lineup. Text-to-image workflow, Release
Sources
- FLUX.2 release announcement
- FLUX.2 dev model card
- Official ComfyUI FLUX.2 dev tutorial
- BFL reference implementation and licenses
- BFL Diffusers deployment configurations
- ComfyUI installation and workflow documentation
- Official ComfyUI Klein tutorial
- Official dev text-to-image workflow and component list
- fal Turbo adapter model card
Top comments (0)