comfyui-mixlab-nodes: turning a ComfyUI workflow into a web app

What comfyui-mixlab-nodes is for

ComfyUI users build graphs, then hand them to someone else who cannot read a graph. This extension attacks that gap. Its headline feature is Workflow-to-APP: an AppInfo node takes a workflow and exposes it as a web app, with the README stating that multiple web apps can be switched between, that a published workflow can be reopened for editing from the right-click menu, and that apps can be assigned categories. The rest of the pack is a grab bag aimed at the same audience: ScreenShareNode and FloatingVideoNode for capturing a screen stream, GPT nodes that talk to ChatGPT, ChatGLM3, ChatGLM4 or a local LLM, speech recognition and speech synthesis nodes, 3D-related nodes, and prompt helpers such as PromptSlide, randomPrompt, ClipInterrogator and ChinesePrompt. It is for people who ship ComfyUI-based tools to non-technical users, and for tinkerers who want screen capture or an LLM call inside a graph rather than beside it. It is not a minimal dependency: pyproject.toml lists numpy, pyOpenSSL, watchdog, opencv-python-headless, matplotlib, openai, simple-lama-inpainting, clip-interrogator, transformers, lark-parser, imageio-ffmpeg, rembg[gpu], omegaconf, Pillow, einops, trimesh, huggingface-hub and scikit-image.

How the Workflow-to-APP mechanism is wired

The AppInfo node is the switch. According to the README, you configure it inside a workflow and the workflow becomes a web app; the app's inputs are not arbitrary graph inputs but a fixed allowlist. Exactly nine node types can appear as interface inputs: Load Image, VHS_LoadVideo, CLIPTextEncode, PromptSlide, TextInput, Color, FloatSlider, IntNumber, CheckpointLoaderSimple and LoraLoader. Outputs are likewise fixed: PreviewImage, SaveImage, ShowTextForGPT, VHS_VideoCombine and PromptImage. Seed handling is centralized, with the README naming SamplerCustom and KSampler as the supported samplers. That list is the real architecture. Anything outside it stays in the graph and is not exposed, which keeps the generated UI simple but also means a workflow built around custom nodes of your own cannot be published as-is. The README also documents an App mode batch prompt feature that groups dynamic prompts and runs them together, and a mode that renders output onto the ComfyUI background in a TouchDesigner-like style. Version 0.46.0 is what pyproject.toml declares, and the newest release listed in the repository is v0.46.0 from 2024-10-14; the last push to the default branch was on 2026-06-04.

Installing comfyui-mixlab-nodes and publishing a first app

The repository carries an install.bat and a requirements.txt, and dependencies are also declared in pyproject.toml, so the expected route is a clone into ComfyUI's custom_nodes directory followed by a pip install of the requirements. The README gives one directory-name constraint that trips people up: the plugin folder must be named exactly comfyui-mixlab-nodes, and a zip download adds a -main suffix that has to be removed. If the web app opens blank, that is the first thing the README tells you to check.

cd ComfyUI/custom_nodes
git clone https://github.com/shadowcz007/comfyui-mixlab-nodes.git
cd comfyui-mixlab-nodes
pip install -r requirements.txt

After restarting ComfyUI, add an AppInfo node to the graph you want to publish. The README's own example is a text-to-image app, and it links a ready-made workflow and a matching app JSON. The inputs you can wire into the app are the nine node types listed earlier, so a minimal graph is a CheckpointLoaderSimple, a CLIPTextEncode, a KSampler with its seed exposed, and a PreviewImage or SaveImage as the output.

# the README points at these files for a working text-to-image app
# workflow/Text-to-Image-app.json
# example/Text-to-Image_3.json

One deployment detail is documented explicitly: if image upload fails over a local network or from a cloud host, serve over HTTPS and use the service on port 8189. The README credits this note to a user report rather than describing it as a general requirement, so treat it as the known workaround for that symptom.

Where the fixed node allowlist bites

The nine-input, five-output allowlist is the honest limitation of Workflow-to-APP. A graph that depends on a node outside those lists cannot be published without restructuring it, and the README does not describe a fallback for exposing other node types. The same silence applies to rollback: there is no documented way to revert a published app to an earlier configuration, and no documented versioning for app JSON files. Speech and vision nodes carry their own weight. The README notes that MiniCPM-V 2.6 int4 runs at about 7GB of GPU memory, and requirements.txt pulls rembg[gpu], bitsandbytes and accelerate, so the pack is not a light addition to an existing environment. It also installs directly from git for two dependencies, SenseVoice-python and openai/swarm, which means a fresh install depends on those repositories being reachable at install time. If you want a small, single-purpose node pack, this is the wrong tool; the pack bundles screen capture, LLM clients, TTS, 3D and image utilities in one install.

Alternatives and how they differ

The most direct comparison is the ComfyUI WAS Node Suite, which also bundles a large set of utility nodes for image, text and logic work. The difference is scope of ambition: WAS Node Suite extends what a graph can do internally, while comfyui-mixlab-nodes adds a publishing layer on top of the graph through AppInfo, plus browser-side capture through ScreenShareNode and FloatingVideoNode. If your goal is a browser front end, the alternative is writing your own UI against ComfyUI's API, which gives you full control over which nodes are exposed and how state is stored, at the cost of building and maintaining that layer yourself. If your goal is only LLM calls inside a graph, a dedicated LLM node pack is a smaller dependency than this one, which also drags in rembg[gpu] and clip-interrogator. The repository's own README points to companion projects by the same author, including comfyui-js-sdk for front-end use of ComfyUI, comfyui-liveportrait, Comfyui-ChatTTS, comfyui-sound-lab and comfyui-ps-plugin, which suggests the intended workflow is a family of packs rather than this one alone.

Licence and maintenance cost

The project is MIT licensed, both in the repository metadata and in pyproject.toml, which permits commercial use and modification provided the licence and copyright notice are kept. That is a permissive position, but the licence covers this pack only; the dependencies it installs carry their own terms, and the pack pulls in rembg, clip-interrogator, bitsandbytes and two git-sourced packages. Anyone redistributing a bundled environment should check those separately. On maintenance: the repository is not archived, and the last push to the default branch was on 2026-06-04. The newest release listed is v0.46.0 from 2024-10-14, and pyproject.toml still declares version 0.46.0, so the version string has not moved with the commits. The README states compatibility with Python 3.11 and torch 2.3.1+cu121 on the latest ComfyUI, which is a specific claim about a moving target; ComfyUI updates can break node packs that pin torch or transformers ranges. Upgrading means re-running the requirements install and re-checking that the git-sourced dependencies still resolve.

Editorial conclusion

Adopt it if you already run ComfyUI and want a browser front end for a workflow without writing your own UI, and if you accept that only the nine documented input nodes and five output nodes map onto that UI. Skip it if you need a stable, narrowly scoped node pack, or if a single dependency such as rembg[gpu] or clip-interrogator==0.6.0 is a problem on your machine. Before committing, confirm on your own install that the plugin directory is named exactly comfyui-mixlab-nodes, that port 8189 serves over HTTPS if you upload images from another machine, and that the version in pyproject.toml (0.46.0) is what you expect, since the newest release listed in the repository is v0.46.0 from 2024-10-14.

Frequently asked questions

Official sources

1. License: MIT
2. MixLabPro/comfyui-mixlab-nodes on GitHub
3. Project website
4. README
5. Releases