Setting up ComfyUI for local image generation

ComfyUI lets Campfire generate images on your own computer. Once it is set up, Campfire can send image requests to ComfyUI for you.

Use ComfyUI if you want local image generation instead of a hosted image provider such as OpenAI, Gemini or Grok.

What You Need

Before you start, make sure you have:

  • A Mac or Windows PC that can run local AI image models.
  • Enough free disk space for image models. One model can be several gigabytes.
  • Time for the first model download. This can take a while on slower internet.

On Apple Silicon Macs, use the native ComfyUI app. Do not use Docker for normal local image generation on macOS. Docker is fine for some server setups, but the native app is the easier and faster path for most Mac users.

Short Version

  1. Install ComfyUI Desktop.
  2. Start ComfyUI and let it finish setup.
  3. Confirm the local ComfyUI server URL.
  4. Download one small checkpoint model for a first connection test.
  5. Add any ComfyUI templates or custom workflows you want to use.
  6. Download the model files required by those workflows.
  7. Restart ComfyUI if you added custom nodes.
  8. Add a ComfyUI account in Campfire with the same server URL.

After that, use Campfire normally for simple image generation. For advanced ComfyUI workflows, set up and test the workflow in ComfyUI first, then let Campfire call the saved workflow.

Install ComfyUI

Install the official ComfyUI Desktop app for your computer. Download it from the official ComfyUI download page:

https://www.comfy.org/download

On macOS, ComfyUI Desktop is the recommended option for Apple Silicon Macs. During first launch:

  1. Open ComfyUI.
  2. Let it finish its first-time setup.
  3. If it asks how to use your Mac's GPU, choose MPS.
  4. Choose a normal local folder for ComfyUI models and user data.
  5. Keep ComfyUI running.

Choose a Good Storage Folder

ComfyUI models are large. During first setup, ComfyUI Desktop asks where to store ComfyUI files such as models, inputs, outputs, custom nodes and saved workflows.

Do not choose a folder that syncs to iCloud, Dropbox, OneDrive or another cloud-sync service. In particular, avoid Documents if your Documents folder syncs to iCloud.

Choose a local folder instead, such as:

/Users/yourname/AI/ComfyUI

or:

/Users/yourname/LocalAI/ComfyUI

After setup, ComfyUI will create folders such as:

ComfyUI/
├── models/
├── input/
├── output/
├── custom_nodes/
└── user/

Put downloaded checkpoint models in:

ComfyUI/models/checkpoints

More advanced models use other model folders too, such as:

ComfyUI/models/diffusion_models
ComfyUI/models/text_encoders
ComfyUI/models/vae
ComfyUI/models/loras
ComfyUI/models/model_patches
ComfyUI/models/clip_vision

If you accidentally chose an iCloud-synced folder, the easiest fix is to start ComfyUI Desktop setup over:

  1. Quit ComfyUI Desktop.
  2. Open this folder:

text ~/Library/Application Support/

  1. Rename or delete the ComfyUI settings folder.
  2. Start ComfyUI Desktop again.
  3. When it asks where to store ComfyUI files, choose a local non-iCloud folder.

Renaming the settings folder is safer than deleting it, because you can restore it if needed.

Confirm the Local Server URL

Campfire talks to ComfyUI through ComfyUI's local HTTP server.

ComfyUI Desktop commonly runs at:

http://127.0.0.1:8000

Classic self-hosted ComfyUI commonly runs at:

http://127.0.0.1:8188

Use the URL that actually opens on your computer. If ComfyUI Desktop is open at http://127.0.0.1:8000, use that same URL in Campfire.

If you need to check or change the port in ComfyUI Desktop:

  1. Open Settings.
  2. Open Server Config.
  3. Find the Network settings.
  4. Keep the host or listen address set to 127.0.0.1 for normal same-computer use.
  5. Check the port.
  6. Restart ComfyUI Desktop after changing server settings.

For normal Campfire use on the same Mac or PC, ComfyUI can stay localhost-only.

If ComfyUI is running on a different computer than Campfire, change the host or listen address to 0.0.0.0 so other devices on your local network can reach it. Only do this on a trusted local network.

If a browser window opens, that is okay. You do not need to press ComfyUI's Run button for simple Campfire image generation.

Add Your First Model

ComfyUI cannot generate images until it has at least one model.

A model is a large file, usually ending in:

.safetensors

Put checkpoint models in ComfyUI's checkpoint folder. It is usually named:

ComfyUI/models/checkpoints

If you use ComfyUI Desktop, use the model folder chosen during ComfyUI setup. Do not put downloaded models inside the app's internal resources.

After you copy a model into the checkpoint folder, restart ComfyUI. This lets ComfyUI find the new model.

Download a Smoke-Test Model

Start with one simple checkpoint model first. This proves that ComfyUI is installed, Campfire can connect to it, and image generation works at all.

For a first test, prefer a single-file checkpoint model that goes in:

ComfyUI/models/checkpoints

These are useful smoke-test models:

Model Download Put it here Notes
Stable Diffusion 1.5 v1-5-pruned-emaonly-fp16.safetensors ComfyUI/models/checkpoints Small and simple. Image quality looks older, but it is good for testing setup.
Flux.1 Schnell FP8 flux1-schnell-fp8.safetensors ComfyUI/models/checkpoints Good modern starter. Schnell is designed to generate in very few steps and is easier to run than the full Flux setup.
Flux.1 Dev FP8 flux1-dev-fp8.safetensors ComfyUI/models/checkpoints Higher quality, heavier, and may have license restrictions. Check the model license before commercial use.
RealVisXL 5.0 RealVisXL_V5.0_fp16.safetensors ComfyUI/models/checkpoints Larger SDXL checkpoint tuned for photorealistic images. Useful for testing whether your local setup can produce camera-like results.

After downloading:

  1. Move the .safetensors file into ComfyUI/models/checkpoints.
  2. Restart ComfyUI.
  3. In Campfire, refresh or re-open the ComfyUI account or model picker if needed.

These checkpoint models are enough for a first Campfire test. More complex workflows may need additional model files in other folders.

Add ComfyUI Workflows

Campfire can use saved ComfyUI workflows. This is the recommended path for advanced local image generation, because the workflow setup stays inside ComfyUI where the custom nodes, template instructions and model links live.

You can use:

  • A workflow you build yourself in ComfyUI.
  • A workflow from ComfyUI's template browser.
  • A workflow shared by another ComfyUI user.

Campfire does not edit the graph. It discovers saved workflows and asks ComfyUI to run them.

Set Up the Workflow in ComfyUI

Do this in ComfyUI first:

  1. Open or create the workflow.
  2. Install any custom nodes requested by the workflow or template.
  3. Download the model files listed by the workflow.
  4. Put each model file in the folder shown by the workflow or template.
  5. For every active Load Image node, select an image from ComfyUI's imported/input image list.
  6. Press Run in ComfyUI.
  7. Confirm the workflow generates an image.

Do not skip the ComfyUI test. If the workflow does not run in ComfyUI, it will not run from Campfire.

Export API Format

ComfyUI has multiple JSON formats. Campfire needs the API prompt format, not the normal editor workflow JSON.

The API format is the format ComfyUI sends to its /prompt endpoint. It usually has top-level node IDs with class_type and inputs.

The editor format usually contains top-level fields such as nodes, links or groups. That format can be opened in ComfyUI, but it cannot be queued through the API.

Use ComfyUI's API-format export/save option. If Campfire says a saved workflow is not in API prompt format, open the workflow in ComfyUI and export it again in API format.

Save the API Workflow

Put the exported API JSON file in ComfyUI's saved workflows folder.

For a ComfyUI root folder such as:

/Users/yourname/AI/ComfyUI

the workflow folder is usually:

/Users/yourname/AI/ComfyUI/user/default/workflows

Example:

/Users/yourname/AI/ComfyUI/user/default/workflows/flux_dev_uso_reference.json

The file name becomes the workflow name Campfire shows. Use short, descriptive names.

Use the Workflow from Campfire

In Campfire, set the image model to:

comfyui-workflows

Then ask CodeBot to list workflows, or request one directly:

/image model=comfyui-workflows:user-flux_dev_uso_reference create a red backpack on a wooden table

You can also use:

/image workflow=user-flux_dev_uso_reference create a red backpack on a wooden table

For advanced workflow details, use the ComfyUI workflow documentation and the workflow notes shown in Campfire.

Reference Images

Reference images must exist on the ComfyUI server before the workflow runs.

When Campfire runs a compatible workflow, it uploads the reference image to ComfyUI. The saved workflow still needs active LoadImage-style nodes that Campfire can patch.

When testing in ComfyUI manually, uploading an image is not always enough. After upload, select the uploaded image from the imported/input image list for the Load Image node. If a node still points at a missing template file, ComfyUI will fail validation.

Model Files for Workflows

ComfyUI templates usually show the model files and folders they need. Follow those instructions exactly.

Common model folders include:

ComfyUI/models/checkpoints
ComfyUI/models/diffusion_models
ComfyUI/models/text_encoders
ComfyUI/models/vae
ComfyUI/models/clip_vision
ComfyUI/models/loras
ComfyUI/models/model_patches
ComfyUI/models/ipadapter

Some Hugging Face downloads may require you to accept the model license while signed in to Hugging Face. If a downloaded file is tiny, it is probably an error page rather than the real model. Open the download link in a browser, accept the license if asked, then download again.

Which Model Should I Download?

Start small. Add one basic model first, just to prove that Campfire can talk to ComfyUI.

If you want Pick this kind of model Why
A quick first test Stable Diffusion 1.5 checkpoint Small, simple, widely supported. Image quality looks older, but it is good for testing setup.
A better everyday model SDXL checkpoint Better quality than SD 1.5 and still reasonably simple to run.
A photoreal local baseline RealVisXL 5.0 SDXL checkpoint Better for real-photo-looking people, places and objects than generic starter checkpoints.
A modern local model Flux.1 Schnell or Flux.1 Dev FP8 checkpoint Good newer option on stronger hardware. The FP8 checkpoint version is simpler because it can use one checkpoint file.
Advanced reference-image workflows Use the model files listed by the ComfyUI workflow/template Reference-image workflows often require several files in different model folders. Follow the template exactly.
Higher quality experiments Larger Flux, Qwen, SDXL or SD3.5 workflows Better results may be possible, but downloads are larger, setup is more fragile, and generation is slower. Check model licenses before commercial use.

For most people, this order works best:

  1. Install one Stable Diffusion 1.5 or Flux checkpoint for the first test.
  2. Add RealVisXL 5.0 or another photoreal SDXL checkpoint when you want real-photo-looking prompt-only results.
  3. Try Flux.1 Schnell or Flux.1 Dev on newer Apple Silicon Macs or PCs with a good GPU.
  4. Add template-specific workflow models only when you have a specific ComfyUI workflow to run.
  5. Try larger models later when you have time and fast internet.

Do not download ten models at once. Start with one test model and one better model.

Add ComfyUI to Campfire

Once ComfyUI is running, add it in Campfire:

  1. Open Campfire|Settings.
  2. Go to CodeBot.
  3. Click Manage & Add Accounts.
  4. Add a ComfyUI account.
  5. Set the URL to the ComfyUI server URL.

For ComfyUI Desktop, this is usually http://127.0.0.1:8000

For classic self-hosted ComfyUI, this is often http://127.0.0.1:8188

If ComfyUI is running on another computer, use that computer's local network address instead. For example:

text http://192.168.1.25:8000

  1. Save the account.

You do not need an API key for local ComfyUI.

Use It From Campfire

Keep ComfyUI running in the background. Then ask CodeBot to generate an image as usual.

For simple model-based generation, Campfire can send the request to the selected ComfyUI image model.

For saved workflows, set the image model to comfyui-workflows and ask CodeBot to use one of the workflows returned by listImageWorkflows().

If ComfyUI is closed, Campfire cannot use it to generate images locally until you start ComfyUI again.

Troubleshooting

Campfire Cannot Connect

If you changed ComfyUI's port, use your custom port instead.

The ComfyUI App Opens, but Campfire Cannot Reach It

ComfyUI Desktop can be open even when the URL entered in Campfire is wrong.

Open Settings|Server Config in ComfyUI Desktop and check the port. If the app is open at http://127.0.0.1:8000, use that same URL in Campfire.

Keep the host set to 127.0.0.1 for normal same-computer use.

ComfyUI Is Running on Another Computer

If ComfyUI is running on a different Mac or PC than Campfire, configure ComfyUI Desktop on that computer to listen on the local network:

  1. Open ComfyUI Desktop on the computer that will generate images.
  2. Open Settings|Server Config.
  3. In the Network settings, change the host or listen address from 127.0.0.1 to 0.0.0.0.
  4. Keep the port as 8000, unless you changed it.
  5. Restart ComfyUI Desktop.
  6. In Campfire, set the ComfyUI account URL to the other computer's local network address.

For example:

http://192.168.1.25:8000

Use the real local IP address of the computer running ComfyUI. Do not expose ComfyUI directly to the public internet.

Campfire Says No Model Is Available

ComfyUI probably does not see a checkpoint model yet.

Check these things:

  • At least one .safetensors checkpoint file is in ComfyUI's checkpoint folder.
  • The model file finished downloading.
  • ComfyUI was restarted after the model was added.

The ComfyUI Web Page Says a Model Is Missing

This means ComfyUI is looking for a model file that is not installed.

Open the workflow or template's model list, download the missing file, and put it in the folder the template lists.

A Workflow Is Not Listed in Campfire

Check these things:

  • The workflow JSON is saved in ComfyUI/user/default/workflows.
  • The workflow JSON is API prompt format, not editor workflow format.
  • The Campfire ComfyUI account points at the same ComfyUI server where the workflow was saved.
  • The image model is set to comfyui-workflows.

The Workflow Fails Validation

ComfyUI was reachable, but rejected the workflow before running it.

Common causes:

  • A required model file is missing.
  • A custom node is missing.
  • A Load Image node points at an image file that has not been uploaded to the ComfyUI server.
  • An image was uploaded but not selected in the Load Image node.
  • The workflow was exported in the wrong JSON format.

Open the workflow in ComfyUI, fix the validation error there, run it successfully, then export the API workflow again.

Generation Is Slow

Local image generation depends on your computer and the model.

Small laptops can work, but generation may take several minutes per image. A newer Mac with a stronger Apple Silicon GPU, or a PC with a supported NVIDIA GPU, will usually be faster. You can also run ComfyUI on a different faster machine, such as a Mac mini or dedicated server hardware, and connect Campfire to it by using that machine's local network URL.

For quick tests, use a smaller image and fewer steps. Once the result looks roughly right, generate a higher-quality version.

See Also