Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-API-DockerCPU

ComfyUI Extension: ComfyUI-API-DockerCPU

Repo Name

ComfyUI-API-DockerCPU

Author
trustypangolin (Account age: 2367 days)
Nodes
View all nodes(1)
Latest Updated
2026-11-09
Github Stars
0.0K

How to Install ComfyUI-API-DockerCPU

Install this extension via the ComfyUI Manager by searching for ComfyUI-API-DockerCPU
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-API-DockerCPU in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-API-DockerCPU Description

ComfyUI-API-DockerCPU connects ComfyUI to cloud AI models for GPU-free image generation, editing and upscaling, with helpers for Hugging Face models, image metadata, file saving and face analysis.

ComfyUI-API-DockerCPU Introduction

ComfyUI-API-DockerCPU brings several cloud-based AI services and practical workflow tools into one ComfyUI extension. You can add nodes for Replicate and fal.ai (https://fal.ai/) models, browse configured Hugging Face repositories, and use tools for organizing images, prompts, faces, and workflow connections.

The extension is designed to work with ComfyUI_API-DockerContainer, which runs ComfyUI on a computer or server without a dedicated GPU. Replicate and fal.ai generation runs through their external APIs, so those tasks do not require a local GPU. The extension also includes local utility and face-analysis nodes; those have their own processing requirements.

For an artist, this means you can build a ComfyUI workflow around cloud image generation, editing, or upscaling, then use the included nodes to name outputs consistently, retain information about source images, and review faces. Cloud services may charge for requests, so use dry-run mode when you only need to check a workflow.

How ComfyUI-API-DockerCPU Works

Think of a ComfyUI workflow as a set of connected workstations. One node supplies a prompt or image, another sends the job to a model, and later nodes inspect or save the result.

  • Replicate and fal.ai nodes turn the settings you choose into an API request. The provider runs the selected model remotely and sends the result back to your workflow.
  • Model-specific controls come from each model’s API description, or schema. As a result, a generation node and an upscaling node can show different inputs appropriate to their tasks.
  • Hugging Face nodes help you select models from configured repositories and prepare LoRA URLs. Their presence does not mean every Hugging Face model runs as a cloud inference job through this extension.
  • Utility nodes handle work around the creative process, such as loading a folder of images, reading saved metadata, and writing outputs to predictable locations.

For example, you could load a photograph, send it to an upscaling node, and save the result under a chosen production stage while carrying forward information about the original. Before making the paid API request, you could enable the node’s debug mode to inspect what it would send.

ComfyUI-API-DockerCPU Features

Cloud Model Nodes

Replicate nodes appear under 🎨 DockerCPU API/Replicate, and fal.ai nodes under 🎨 DockerCPU API/FalAi. Each supported model has a node with controls based on that model’s available parameters. Depending on the model, nodes can accept or return images, audio, or text. They also provide a JSON payload output so you can inspect the request.

Debug/dry-run mode is available on schema-based nodes. Set debug to True to skip the API call and output the payload that would have been sent. If an input image is available, the node returns the first one as a mock image output. This helps you check connections and settings without paying for that test request; it does not show what the model would actually create.

You need the relevant provider credential for live requests: REPLICATE_API_TOKEN for Replicate or FAL_KEY for fal.ai.

Hugging Face Helpers

Under 🤗 HuggingFace, the HF Model Selector lets you browse safetensor models in configured repositories. HF Lora URL Builder creates a Hugging Face file URL from a repository ID and filename, while HF Lora Combiner collects several LoRA URLs into an array for another node. A LoRA is an add-on that can steer a compatible model toward a particular look or subject; which combinations work depends on the downstream model and workflow.

Image, Prompt, and File Utilities

These nodes appear under 🎨 DockerCPU API/Utilities.

  • Load Image With Metadata loads an image and exposes its image, mask, dimensions, filename information, hash, and available embedded prompt or workflow metadata. A crop variant also exists.
  • Load Images From Folder With Metadata loads images from an input subfolder in natural filename order, so img2 comes before img10. You can filter by filename pattern, include subfolders, and choose a starting point and limit. Images are returned at their native sizes.
  • Load Prompt From File lets you pick a .txt prompt under input/prompts/, including its subfolders. It can reread the file when its contents change, which is useful while refining a written prompt outside ComfyUI. Press R in ComfyUI to refresh the file list.
  • Save Image With Filename saves a PNG using a chosen stage, mirrored directory, source filename stem, and suffix. For a batch, it adds numbered suffixes. Its if_exists setting can stop on an existing file, overwrite it, or skip it.
  • Save Text With Filename saves text using the source image’s filename or the same stage-based naming pattern. For example, you can keep a caption alongside an image with a matching stem.
  • Load ZIP Archive loads PNG images from a ZIP file URL or path.
  • Load Replicate Prediction retrieves the output of a Replicate prediction by its ID.
  • Save AMD Video provides a video-saving path with an AMD GPU audio fix.

The image loaders and Save Image With Filename can also carry provenance: information identifying the original library image from which an output derives. If an original can be resolved, ORIGINAL gives its library path and ORIGINAL_SHA256 gives its file hash. You can connect those outputs to the saver so the information travels with the PNG. Turn on require_original when an image must not be saved unless both original-image details are available. If the library is unavailable, provenance may be unknown; do not treat an empty value as a verified match.

Face Analysis and Sorting

Face nodes appear under 🎨 DockerCPU API/Face. They run on CPU and use a GPU when one is available. The InsightFace-based nodes use ONNX models, rather than the insightface Python package.

  • Face Analyze (InsightFace ONNX) detects faces and returns their positions, scores, count, and analysis data. Sort faces left to right or by detection score; the chosen order stays consistent across its outputs.
  • Draw Numbered BBoxes draws numbered boxes around detected faces. You can change the ordering, label starting number, line width, label size, color, and whether scores appear.
  • Face Similarity (ArcFace) compares one face with reference analysis and/or a folder of reference images and embeddings, then returns the best match and a ranking.
  • Face Identify (ArcFace) compares each face in an image with labeled reference folders. Adjust threshold to require a stronger match and margin to require a clearer lead over the runner-up; faces that do not qualify receive the chosen unknown label. Its per-face crops and filenames can feed a review or saving workflow.
  • Face Cluster (ArcFace) groups faces from a folder by similarity. Changing distance_threshold changes how readily faces join the same group, making this useful for preparing reference folders that you can review.
  • Collect Labelled Faces records a review made by moving face crops into labeled folders and can optionally collect confirmed crops as references.
  • Load Face Detection Model (MediaPipe, device) loads a MediaPipe face landmarker for use with native MediaPipe detection and mask nodes.

Face matching is an aid to organizing and reviewing images, not a guarantee of identity. Check uncertain matches yourself, especially before using labels in a final collection.

Canvas and Touch Controls

  • Touch and Apple Pencil usability: On supported touch devices, menu entries are easier to tap, and pencil taps on dropdowns and the canvas are handled more reliably. Settings → DockerCPU → Pencil → Ignore finger touches lets you rest your hand on the screen while using a pencil. After the first pencil tap, a floating pencil toggle can switch that setting; a finger can still tap the toggle. Apple Pencil double-tap and squeeze are not supported through the browser.
  • Link Chain Highlight: Tap or click a wire’s body to highlight the connected chain of the same data type, including reroutes and the nodes along the way. Tap it again, tap empty canvas, or press Esc to clear the highlight. Turn it off under Settings → DockerCPU → Links. In Nodes 2.0, wires animate but node outlines are omitted.
  • Zip Ties: Bundle several wires into a movable ribbon to make a busy canvas easier to read. Hold Z and drag across wires, or use Tie wires from the empty-canvas context menu if you do not have a keyboard. You can move, rotate, recolor, or remove ties. Settings → DockerCPU → Zip Ties controls whether they appear, the color of new ties, and the gesture key. Ties are saved with the workflow; workflows still draw their wires normally without the extension.

ComfyUI-API-DockerCPU Models

The original README lists the following supported model IDs. Choose a model by the job you want to do, then check its node for the inputs it accepts. Availability and provider-side behavior can change.

Replicate

ModelWhat to use it for
black-forest-labs/flux-2-klein-9bFast text-to-image generation.
black-forest-labs/flux-2-klein-9b-base-loraFast FLUX generation using a compatible LoRA.
black-forest-labs/flux-2-maxImage generation when fidelity is the priority.
black-forest-labs/flux-2-proProfessional image-generation work.
google/nano-bananaFast text-to-image generation.
google/nano-banana-proA Pro image-generation option with more images.
google/nano-banana-2An extended Nano Banana image-generation option.
openai/gpt-5Image understanding and image-to-text tasks, such as captions.
lucataco/qwen3-vl-8b-instructVision-and-language tasks involving images and text.
mattsays/sam3-imagePrompt-based image segmentation, such as isolating a subject.
papina/seedvr2Enhancing and upscaling images.
philz1337x/crystal-upscalerHigh-quality image upscaling.
topazlabs/image-upscaleImage upscaling.
sczhou/codeformerRestoring or enhancing faces.
zsxkib/seedvr2Upscaling video.
zsxkib/realistic-voice-cloningAudio processing and AI song covers using RVC voice models.
prunaai/z-image-turbo-loraFast image generation with LoRA support.
tencentarc/gfpganFace restoration.

For example, choose an upscaler when you already have an image and want a larger or enhanced version; choose a segmentation model when you need to separate a subject from an image. A face-restoration model serves a different purpose from a general upscaler, so compare results on the kind of source material you actually use.

fal.ai

ModelWhat to use it for
fal-ai/flux-2/klein/9b/editEditing an existing image with FLUX.2 Klein.
fal-ai/seedvr/upscale/imageUpscaling an existing image with SeedVR.

Configured Hugging Face Repositories

These are repositories exposed through the Hugging Face helpers, not a list of additional cloud inference nodes.

RepositoryListed purpose
renderartist/Technically-Color-Z-Image-TurboColorization and style transfer.
starsfriday/FLUX.2-klein-AC-Style-LORAAn anime/comic-style LoRA for FLUX.
Papina/Lora-CharactersCharacter-style LoRAs.

Troubleshooting ComfyUI-API-DockerCPU

A live Replicate or fal.ai node cannot make a request.

  1. Check which provider the node uses.
  2. Confirm that its credential is available to ComfyUI: REPLICATE_API_TOKEN for Replicate or FAL_KEY for fal.ai.
  3. Check that the chosen model’s required node inputs are filled in.
  4. Run once in debug mode to inspect the request payload without calling the provider. If the payload looks right but the live request still fails, check the provider’s response or ComfyUI logs for the specific error.

The node returns my input image instead of a newly generated result. Check whether debug is set to True. In dry-run mode, returning the first input image is expected. Set it to False when you are ready to make a live request, which may incur a provider charge.

An expected Hugging Face model does not appear. The selector browses configured repositories, not the entire Hugging Face Hub. Confirm that the repository you need is among those configured for this extension. A token is optional for public repositories but required for private ones.

A saved image has no original-library information.

  1. Check the loader’s ORIGINAL and ORIGINAL_SHA256 outputs.
  2. Connect both outputs to the corresponding Save Image With Filename inputs.
  3. If either is empty, check whether the source image carries provenance metadata or whether the original library is accessible to the workflow.
  4. Enable require_original if you would rather stop saving than produce an image without a resolved original.

A file was not saved where I expected. For Save Image With Filename, check stage, mirror_dir, stem, and suffix: together they determine the location under the output folder. Check if_exists if the name is already taken. The saver refuses paths outside the output folder or under the input folder.

Face identification returns an unknown label. Check that the reference folder contains examples under the intended label and review the threshold and margin settings. Requiring a higher score or a larger lead over the next label makes assignments more selective. Review the crop and scores before relaxing either setting.

A pencil tap behaves like a drag, or finger input is blocked. Check the DockerCPU Pencil settings. Ignore finger touches intentionally blocks finger input on the canvas while enabled; use the floating pencil toggle to switch it off. The touch improvements are intended for touch devices, while desktop mouse and trackpad use is unaffected.

Frequently Asked Questions

Do I need a dedicated GPU? Not for inference sent to the supported cloud API nodes. Some other nodes do local work; the face nodes can run on CPU and use a GPU when available.

Will debug mode generate a free preview? No. It skips the API request, shows the payload, and can pass through the first input image as a mock output. It does not generate a model result.

Are models downloaded to my computer? The extension’s DockerCPU cloud API design does not require local downloads of its Replicate or fal.ai inference models. Local face-analysis nodes have separate model requirements.

Can I keep source details with a finished PNG? Yes. Save Image With Filename embeds prompt and workflow information and writes a restore metadata entry. Connect the loader’s original-path and hash outputs when you want available provenance carried forward.

Will zip ties break a workflow opened without this extension? The README states that its wires display normally without the extension. Zip-tie layout information remains in the workflow data for use when the feature is available again.

Learn More about ComfyUI-API-DockerCPU

ComfyUI-API-DockerCPU Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-API-DockerCPU detailed guide | ComfyUI