What ComfyUI actually is
The ComfyUI docs describe it as an open-source, node-based application for generative AI, covering image, video, audio and 3D (docs.comfy.org). In plain terms, each node is one step: load a model, add a prompt, apply a style, resize the result. You wire the nodes together on a canvas and the image comes out the other end.
That wiring is called a workflow. It's saved as a small JSON file, and it can also be tucked inside a generated image, so a colleague can drag the picture into ComfyUI and get the whole setup back. Once a workflow works, anyone on the team can reuse it with new inputs.
Why a custom workflow beats starting from scratch
Typing prompts into a chat tool is fine for one-off ideas. When a team needs fifty images in the same look for a campaign, it gets messy fast: different settings, different results, nobody sure which version was approved.
A workflow fixes the recipe. The model, the style, the image size and the finishing steps stay the same every time, and people only change the parts they're meant to change, like the product photo or the background brief.
Workflows we build
Every team's needs differ, so these are starting points we shape around your briefs.
- Product shots: your real product photo placed on new backgrounds, with the product itself kept as it is
- Consistent characters or mascots across many images, using LoRA training or reference-image methods
- Upscaling and clean-up to prepare images for print and large banners
- Batch variations: one design in several sizes, colours or seasonal themes
- Image-to-video: short clips from a still, using an open model such as Wan 2.2
Running workflows automatically
A self-hosted ComfyUI server can take jobs over its own API. Another tool sends it a workflow and an input image, and the result comes back without anyone opening the canvas (ComfyUI server routes). That means a spreadsheet of products, a form on your intranet or an automation tool can queue the work for you.
We can wire this up as part of the job, or connect it to a wider business automation so finished images land in a shared folder for a person to check.
Local machine or cloud GPU?
ComfyUI runs best on a strong graphics card. If your studio already has a capable workstation, running locally keeps files in-house and has no per-image fee. The trade-off is upkeep: models and custom nodes change often, and someone has to keep things updated.
Cloud GPUs avoid buying hardware. You rent the power when you need it. As of October 2026 Comfy also has its own cloud API on paid tiers (Comfy Cloud API), though hosted options change quickly. We talk through how much you'll use it before recommending either route.
Model licences matter
Each AI model comes with its own licence, and they differ a lot. Some, like Qwen-Image, use the permissive Apache 2.0 licence. Others restrict commercial use or need a separate agreement. Before a workflow goes into client work, we check the licence of every model in it and tell you what we found.
The final call on commercial use stays with you, so keep a note of which models each workflow uses.
Handover and training
A workflow is only useful if your team is comfortable with it. We label the nodes people should touch, hide the ones they shouldn't, and walk through it with you on a call. You keep the workflow files.
Tell us about your briefs at /start/ and we reply within 1 working day with questions and a quote.