What the assistant is for
TokenByte built the ComfyUI Assistant Lite to solve a real beginner problem: people open ComfyUI, see a blank graph, and do not know which nodes belong in the workflow. The assistant asks what you are trying to make, your VRAM class, model lane, and control needs, then returns a practical node order and first test plan.
The model lane is a planning choice, not a bundled download. Users can choose from SDXL, Stable Diffusion 3.5, FLUX.1 Schnell, FLUX.1 Dev or Kontext-style workflows, SD 1.5, Pony/anime checkpoints, Qwen-Image, HiDream, Hunyuan, Wan video/image-to-video, or upscaler-only workflows. They still need to download any models themselves from official or trusted sources and follow each license.
Keep baseline graphs, experiment branches, and output notes together.
Capture model, seed, resolution, sampler, steps, and hardware notes for comparisons.
Separate workflow problems from true GPU, VRAM, storage, or cooling bottlenecks.
Model choices included
| Lane | Best use | What the assistant does | What it does not include |
|---|---|---|---|
| SDXL / SDXL Turbo | General ComfyUI learning, portraits, products, styles, and baseline image tests. | Plans the classic checkpoint, prompt, sampler, VAE, and save-image flow. | No SDXL checkpoint files. |
| Stable Diffusion 3.5 | Newer Stable Diffusion workflows, prompt following, and composition tests. | Calls out that SD 3.5 workflows may need different loader/text encoder patterns than SDXL. | No Stability model files. |
| FLUX.1 Schnell / Dev | Higher-quality FLUX-style workflows and users comparing current model families. | Suggests FLUX-specific loader, text encoder, guidance, sampler, and VAE expectations. | No FLUX weights or license bypass. |
| SD 1.5 / Pony / anime | Low-VRAM learning, ControlNet practice, character/anime checkpoint workflows. | Keeps the graph lightweight and warns about prompt format and model rules. | No private checkpoints or LoRAs. |
| Qwen-Image / HiDream / Hunyuan | Advanced model comparison and newer image-model experiments. | Treats these as advanced lanes with compatibility and license checks. | No third-party model assets. |
| Wan video / upscaler-only | Video, image-to-video, or post-processing workflows after still-image baselines work. | Warns about VRAM, RAM, storage, and runtime pressure. | No video models or upscale weights. |
How to install and use it
- Download the ZIP.
- Unzip the folder anywhere on macOS, Linux, or Windows.
- Open
index.htmlin your browser. - Pick your goal, VRAM, model family, and control needs.
- Copy the generated node plan into your ComfyUI notes and build the graph manually.
This is intentionally not bundled as a native Mac app yet. The current version is a safe local browser app so it can work across Mac, Linux, and Windows without installing dependencies. A native package can come later after the GitHub repo is public and tested.
Privacy checks before release
Public package gate
- No personal name, home path, machine username, address, phone, or personal email in examples.
- No API keys, passwords, tokens, cookies, local database files, or private browser/session data.
- No model files, LoRAs, checkpoints, embeddings, VAEs, upscalers, or third-party assets bundled.
- No hardcoded absolute paths. Examples should use placeholders like
/path/to/ComfyUI.
Download access
The Lite ZIP is available now. It is a small local browser helper, not a ComfyUI custom node package and not a model bundle.
Download TokenByte ComfyUI Assistant Lite
Unzip it, open index.html, and use it as a workflow planning companion while building nodes in ComfyUI.
GitHub repo
The source package is ready now as a ZIP. When the public GitHub repo is created, it should use the same shape as a normal open-source project: app source, README, install docs, usage docs, license, and security policy.
Source package
Includes README.md, SECURITY.md, LICENSE, install docs, usage docs, and the local browser app.
Release roadmap
| Step | What happens | Visitor impact |
|---|---|---|
| 1. Lite ZIP | Local browser assistant for workflow planning. | Live now. |
| 2. GitHub | Publish source and issues when the public repo is created. | Gives the tool a shareable developer home. |
| 3. Native app | Evaluate Mac/Linux packaging after users test the Lite build. | Better install experience if demand is real. |
ComfyUI Assistant FAQ
Is the assistant downloadable right now?
Yes. The Lite ZIP is available as a direct download. It runs locally by opening index.html in a browser.
Do users need to already have the models?
Yes. The assistant gives workflow planning choices, but it does not ship models. Users need to download model files, VAEs, LoRAs, or upscalers themselves from official or trusted sources and follow each license.
Will it expose my private ComfyUI setup?
No. The download does not include private model paths, usernames, secrets, models, LoRAs, checkpoints, or output files.
What should I read first?
Start with the ComfyUI Starter Workflow page, then use the assistant to keep your workflow evidence organized.