Connect explicit processing steps
Use the graph to identify which operation loads a model, consumes a prompt or reference, transforms an intermediate result, and saves the deliverable.
ComfyUI
ComfyUI turns generative-media steps into reusable node graphs. Explore workflow JSON, model and custom-node dependencies, queued execution, and API handoffs to design a visual-production process with inspectable inputs and outputs.
Prepare a repeatable product-image workflow
Bring the media brief and the workflow or assets already available. The request can map generation and finishing stages, identify node and model dependencies, and define representative output checks before increasing batches or transferring the production recipe.
Describe inputs, background/style constraints, model availability, hardware, and output size.
Copy the request into a new Floatboat conversation and provide the required material.
Download FloatboatYour request for Floatboat
Preview and copy the request. It is not sent automatically from this page.
Reference: ComfyUI https://github.com/Comfy-Org/ComfyUI Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding. My task: Use ComfyUI’s official workflow documentation to plan consistent backgrounds for ten product photos. I will provide the originals, target size, and available hardware. Identify a suitable graph, required models, and a two-image validation step before installation or batch execution. Expected deliverable: A proposed node workflow with required models, input/output mapping, and a small image-validation plan.
About ComfyUI
ComfyUI is a node-based interface and execution engine for generative-media workflows. A graph connects operations such as loading a model, conditioning generation, processing media, and saving an output. The open core is maintained in Comfy-Org/ComfyUI; desktop packaging and the frontend have related repositories rather than being separate Skills.
Workflows can be saved as JSON, reused, extended with custom nodes, and invoked through API entry points. That makes the graph a concrete description of a production process, but reproducing a result also requires the relevant model files, node versions, settings, and runtime environment.
For work organized in Floatboat, study how a prepared brief becomes a bounded media-generation step and how its outputs return for review. Local execution, paid cloud services, and partner-model API nodes have different requirements. The core license does not grant rights to every model, extension, or generated asset.
Build a Generation Process You Can Inspect
ComfyUI graphs connect explicit media operations with reusable workflow state and bounded jobs. Compare how inputs, model dependencies, and intermediate outputs remain attached to each stage, making it easier to trace a weak result and hand off a reproducible recipe.
Preserve a reusable workflow
Keep workflow JSON alongside the model identifiers, custom-node requirements, and relevant settings needed to reproduce the task.
Hand off a bounded job
Use the documented queue and API surfaces when another application prepares inputs or collects results. Define the actual integration separately.
Use Cases for ComfyUI
Image series, finishing stages, and portable production recipes need both media and environment checks. Explore how ComfyUI graphs preserve operation order, model dependencies, and representative inputs so results can be reviewed and reproduced by another collaborator.
Prepare a consistent image series
A consistent image series starts with product originals, style constraints, output sizes, and available models. Keep the workflow and model setup stable while varying the brief or source asset, then compare every result against the same subject and background criteria. Retain the inputs so the series remains traceable.
Separate generation from finishing
A media graph can separate generation from masking, resizing, upscaling, and compositing. Represent the required finishing stages explicitly and keep each result connected to the operation that produced it. When a frame looks wrong, the graph should help locate the responsible stage instead of forcing the entire workflow to be rebuilt.
Reuse a reviewed production recipe
A reviewed workflow JSON is only part of a reusable production recipe. Package the custom nodes, model dependencies, representative inputs, and acceptance images before handing it to another machine or collaborator. Those materials make the expected output and the environment needed to reproduce it visible together.
Review dependencies before adopting a graph
A shared workflow may rely on custom nodes, model weights, or external services that are not present in the receiving environment. Inspect those requirements and their stated terms before running the graph, then test a representative input under the intended setup. Keep dependency and output checks together so a portable workflow file is not mistaken for a self-contained production environment.
Check ComfyUI models and dependencies
Hardware and operating-system support are only part of a ComfyUI setup. Review custom-node dependencies, model terms, and any external API conditions alongside the workflow, since each can affect whether another environment can reproduce the intended output.
Choose a supported runtime
Follow the official installation path for your operating system and hardware. Memory and performance depend on the selected models and workflow.
Review extensions and model terms
Custom nodes add executable dependencies, while model weights and external APIs have their own licenses or service conditions.
Validate a ComfyUI workflow before batching
A reusable media workflow should first succeed with representative inputs and known dependencies. Resolve the environment, inspect saved outputs, and retain the setup needed to reproduce them before increasing the batch size or handing the recipe to another machine.
Test a complete sample
Load the workflow, resolve its dependencies, run a representative input, and inspect the saved result before automating a batch.
Related Software Projects
Compare the workflows, operating requirements, and source licenses of these projects.
ComfyUI FAQ
Is ComfyUI a Skill?
No. It is a visual application and execution engine. A separate Skill may explain how to operate it, but the node graph and runtime are software components.
Is a workflow JSON enough to reproduce an output?
Not by itself. Model files, custom nodes, versions, seeds, settings, and hardware-dependent behavior can affect execution and results.
Are custom nodes included in every installation?
No. A shared workflow may depend on extensions that must be installed and reviewed separately. Check the graph's required node types before running it.
Can an agent invoke ComfyUI through an API?
ComfyUI documents API-based workflow execution. A calling agent still needs an implemented connection, access to the deployment, validated inputs, and a way to review outputs.
Does local ComfyUI require a paid cloud plan?
The open core can run locally with suitable dependencies and hardware. Official cloud services and partner-model API nodes are separate offerings.
Does the core license cover all models and media?
No. The core uses GPL-3.0; model weights, custom nodes, provider APIs, and input assets can have separate conditions.