A ComfyUI workflow is a saved graph of steps: load a model, take a prompt or a photo, apply the changes you choose, and write out the finished image. For a business team the point is repeatability. Build the workflow once, save it as a file, and everyone produces images the same way, on a private server that keeps the source photos in-house.
The problem most teams hit with AI images
Most marketing and product teams we meet have already made AI images. One person got a good result once and could not repeat it. Three people used three different tools and three different prompts, and nothing matched. Along the way, the original product photos and a few unreleased designs were uploaded to cloud services nobody had vetted.
The fix is rarely a better prompt. It is a workflow that the whole team runs the same way, on the same models, with the same style settings. That is what ComfyUI was built for, and it is why we run it as the image engine inside SecondBrain. If your images keep drifting, this explains why.
What a ComfyUI workflow actually is
ComfyUI shows image generation as a graph. Each box is a node that does one job: load a model, read a prompt, load a reference photo, mask an area, sample, upscale, save. Links between the boxes carry the data. The whole graph is a workflow, and it saves as a small JSON file you can reopen, copy and share.
Two details matter for teams. First, ComfyUI also embeds the workflow inside every PNG it saves, so dragging an old output back into the editor restores the exact settings that made it. Second, the software ships with a template library and the official documentation lists working examples for every supported model, so a team never starts from a blank canvas.
Where does ComfyUI keep workflows? In the workflows folder under its user directory, one JSON file per saved workflow, plus the copy embedded in each image. Back up that folder and you have backed up the team’s method.
The four workflows a product or marketing team needs
Most of the work a business asks for comes down to four graphs. Build these well and the rest are variations.
1. Clean product shot to scene variations
Photograph each product once on a plain background. The workflow masks the product so its pixels stay locked, then an edit model such as Qwen Image Edit or Flux 2 Klein places it in the scenes you describe: a kitchen counter, a festive table, a studio gradient. One photo becomes a season of imagery, and the product never changes shape.
2. Brand style transfer
A style reference image plus a fixed prompt template gives every generation the same lighting, palette and finish. This is the workflow that stops the “three people, three looks” problem. The prompt template lives in a shared library, not in someone’s head.
3. Posters, banners and infographics with real text
Older image models mangled words. Qwen Image renders text reliably and outputs at native 2K, so promotional posters, price boards and event banners can come out of the same pipeline with the copy already in place. We cover the model choice in Qwen Image and Z-Image in ComfyUI.
4. Upscale and clean-up for print and web
A short graph that takes any approved image, upscales it, removes artefacts and exports the sizes your channels need. Run it as a batch at the end of the day rather than one image at a time.
Where to get workflows, and how to share them across a team
Start with the built-in templates and the official examples, then adapt. Community workflows are plentiful but often depend on custom nodes that break on the next update, so we treat them as ideas rather than production tools.
For a business, the workflows themselves are only half the method. The other half is the prompts and styles that feed them. SecondBrain keeps those in a prompt library grouped by campaign, with curated styles preloaded, so a new hire picks a style and a prompt from the menu and gets the house look on day one. The ComfyUI graph runs underneath and most staff never see a node.
One more rule: run everything on one server. ComfyUI releases updates every few days, and models change just as fast. When the whole team generates on the same machine, everyone is on the same version, the same models and the same custom nodes, and yesterday’s workflow still works today.
Hardware and where it runs
You do not need a server room. Z-Image Turbo and Flux 2 Klein 4B both fit a 16GB workstation GPU, which is what our standard private AI server ships with. Heavier models and video work suit the larger plan. The server sits in your office or a rack, and the team reaches it through a browser. If you are weighing that against renting GPU time, read ComfyUI cloud or a private server.
Where workflows are not the answer
A workflow pays when the same kind of image is needed again and again. For a one-off visual, a single prompt in a chat tool is faster and nobody should build a graph for it.
Some images should not be generated at all. Where a photo carries a legal or safety claim, a regulated product, a medical device, a food label, the real photograph stays. Generated imagery belongs in campaigns, concepts and variations, with a person approving what goes out.
Finally, a workflow needs an owner. If nobody on the team is responsible for the style library and the server, the method drifts within a month. We will say so before we build anything. For the wider picture on what the tool is and is not, see ComfyUI for business use.
MT Labs helps companies across Singapore deploy AI tools they actually own. Private infrastructure, no recurring cloud subscriptions, and a setup built around how your team already works. Whether you need one product-image workflow or a full image pipeline for the marketing team, we build it on a server you own and train your people on it. Get in touch and we’ll start with the images you make most often.
FAQ
What is a ComfyUI workflow?
A ComfyUI workflow is a saved graph of nodes that describes how an image is made: which model to load, what prompt or reference photo to use, which edits to apply, and how to save the result. It is stored as a JSON file and can be reopened, copied and shared, so a team can produce images the same way every time.
Where does ComfyUI keep workflows?
Saved workflows sit in the workflows folder inside ComfyUI's user directory, one JSON file each. ComfyUI also embeds the workflow inside every PNG it generates, so dragging an old output into the editor restores the settings that produced it.
Where can I get ComfyUI workflows?
ComfyUI ships with a template library, and the official documentation publishes working examples for each supported model. Community workflows are widely shared, but they often rely on custom nodes that break after updates, so businesses are better off adapting the official examples into a small internal set.
Which ComfyUI workflows are best for product images?
Four cover most business needs: a clean product shot placed into scene variations with the product masked, a brand style transfer using a reference image and a fixed prompt template, a text-capable poster and banner workflow using Qwen Image, and a batch upscale and clean-up workflow for final export.
Can a non-technical marketer use ComfyUI workflows?
Yes, when the graphs are wrapped in a simpler front end. In SecondBrain, staff choose a style and a prompt from a shared library and the ComfyUI workflow runs underneath, so most of the team never sees a node. Building and maintaining the graphs is the technical part, and that is what we set up.
Do we need our own server to run ComfyUI workflows?
ComfyUI runs on a single workstation-class GPU. A 16GB card handles models such as Z-Image Turbo and Flux 2 Klein 4B. A private server in your office keeps product photos and unreleased designs in-house and lets the whole team generate without usage caps, which is the setup we recommend for teams that make images daily.