Image editingComfyUI-ready repository

Comfy-Org

Run flux2-dev

Edit source images with a validated ComfyUI workflow family. Estimated setup is about 30 minutes, with no local installation. From $2.78/hour.

Review live offers See what it costs

Validated runtime

Architecture and task checked

Immutable revision

ed33133cd5

Dedicated workload

Isolated model endpoint

You control runtime

Runs until you stop it

The outcome

What it does

flux2-dev is a validated image editing model.

GPU Vault admits only generated-registry repositories whose exact component paths and hashes match a pinned official ComfyUI core-node workflow.

The model launches in a persistent authenticated ComfyUI workspace. Download outputs before stopping the workload because workspace files are deleted during teardown.

Example workflows

What you can build

Creative production

Build and refine media with an editable node graph.

Workflow iteration

Adjust the bundled starter workflow for your project.

Direct API

Queue prompts through the authenticated ComfyUI API.

Results depend on the model, prompt, sampling settings, and publisher guidance.

Recommended hardware

Your setup

We run this on

1× GPU · 80 GB VRAM

This is the smallest supported topology with enough memory for the model weights, runtime overhead, and the validated context window. GPU Vault also reserves 110 GB for the immutable model files and engine data.

Why this setup?

The compatibility classifier uses safetensors metadata, engine overhead, context memory, and tensor-parallel divisibility. It rejects unsupported architectures and models that need more than eight GPUs.

GPUs: 1
VRAM each: 80 GB
Disk: 110 GB
Tensor parallel: 1
Weight memory: 69.0 GB
Setup estimate: ~30 min

Transparent usage

What will it cost?

Current public starting rate

$2.78/hour

The dashboard shows your exact role-based rate and live offer before approval.

Runs until you stop it

There is no fixed serving duration. Billing follows active compute time from instance creation through teardown.

Approve before launch

Review the exact GPU, base compute rate, setup estimate, and initial wallet hold. No credits are reserved from this public page.

Choose your workflow

Two ways to use it

Open the ComfyUI workspace

Start from the GPU Vault core-node workflow and keep full graph control.

Launch for the console

Available after the model passes its readiness check.

Connect your application

Use the short-lived direct ComfyUI API credential from the workload page.

Launch for API access

Rotate or revoke the workload key at any time.

Exact profile

Model facts

Show details

Publisher: Comfy-Org
Category: image editing
Architecture: ComfyUI:image_flux2
Parameters: Not declared
Context window: Not applicable
Weight memory: 69.0 GB
Repository storage: 69.0 GB
Tensor parallelism: 1
License: other
Exact revision: ed33133cd56476eac818c0943b6f9419b3e4a3a1

View original repository on Hugging Face

Keep exploring

Find another model

More image editing models

Compare other launchable models with the same serving task.

Explore

More from Comfy-Org

Search validated models published by the same organization.

Explore

Browse every launchable model

Explore text generation and pinned ComfyUI image, editing, and video models.

Explore

From Comfy-Org

Publisher documentation

Read model card

FLUX.2-dev

Repackaged model files for ComfyUI.

Original model repository: https://huggingface.co/black-forest-labs/FLUX.2-dev

Place the files in the following folders:

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── flux2_dev_fp8mixed.safetensors
│   ├── 📂 loras/
│   │   ├── Flux2TurboComfyv2.safetensors
│   │   └── Flux_2-Turbo-LoRA_comfyui.safetensors
│   ├── 📂 text_encoders/
│   │   ├── mistral_3_small_flux2_bf16.safetensors
│   │   ├── mistral_3_small_flux2_fp4_mixed.safetensors
│   │   └── mistral_3_small_flux2_fp8.safetensors
│   ├── 📂 vae/
│   │   └── flux2-vae.safetensors

These are quantized weights. The original FLUX.2 repo is here: FLUX.2-dev.

Flux2TurboComfyv2.safetensors by ByteZSzn
Flux_2-Turbo-LoRA_comfyui.safetensors by ByteZSzn

Workflows

Common questions

Before you launch

What happens to image editing outputs?+

Outputs stay on the GPU instance. Download anything you need before Stop or teardown; GPU Vault does not copy catalog workspace files into Supabase storage.

Do I need to install or configure the model?+

No. GPU Vault uses a fixed server-owned runtime configuration and the model’s immutable revision. Select a live GPU offer, explicitly approve the launch, and wait for the readiness check.

How does billing work?+

The workspace runs until you stop it, with the approved rate protected by your initial and rolling wallet holds. If the runtime image cannot download before ComfyUI becomes ready, GPU Vault records zero customer usage and returns the complete hold after confirmed teardown.

Can the model or price change after I approve?+

Approval revalidates the exact model profile, GPU offer, rate, wallet, and operational gates. If a customer-visible model or offer changes before launch, GPU Vault requires a fresh review.

Are my prompts and outputs stored?+

GPU Vault records only endpoint, status, latency, and byte-count metrics for inference requests. Prompt content, documents, embeddings, and generated output are not persisted by the inference proxy.

Ready when you are

Run flux2-dev on a dedicated GPU workload.

Choose an exact live offer, review the rate and wallet hold, then approve the launch from your dashboard.

Launch this model

📂 ComfyUI/ ├── 📂 models/ │ ├── 📂 diffusion_models/ │ │ └── flux2_dev_fp8mixed.safetensors │ ├── 📂 loras/ │ │ ├── Flux2TurboComfyv2.safetensors │ │ └── Flux_2-Turbo-LoRA_comfyui.safetensors │ ├── 📂 text_encoders/ │ │ ├── mistral_3_small_flux2_bf16.safetensors │ │ ├── mistral_3_small_flux2_fp4_mixed.safetensors │ │ └── mistral_3_small_flux2_fp8.safetensors │ ├── 📂 vae/ │ │ └── flux2-vae.safetensors