You found a workflow on Civitai that gives exactly the look you want, or your team built one that works on a single machine. Now it has to produce 2,000 images or 300 clips, and the setup alone has already eaten a weekend: red nodes, missing models, version conflicts. GoraGen is a done-for-you ComfyUI workflow service. People on our team do the ComfyUI workflow setup, agree a test batch with you and then generate the volume on our GPUs. You get reviewed results, not a server to look after.
If you were searching for a ComfyUI workflow developer, an expert or a freelancer to get a workflow running, this is the same job with one difference: at the end you receive the finished batch, not advice and not a configured machine you then have to operate yourself. The consulting part is still there. Before anything runs, you know what the workflow needs, what is closed or restricted, and roughly what it will cost.
Which workflows we take
Any ComfyUI workflow you can share as a link or a file:
- Civitai. A workflow page or a model page with an attached workflow. If you want to run a Civitai workflow you found, the link is enough for us to start the review.
- OpenArt. A link to the workflow page.
- GitHub. A repository with workflows, for example the examples folder of a node pack.
- Your own
.json. The regular workflow export is better than the API format: it keeps notes, groups and, in recent ComfyUI versions, which node pack each node came from. - A PNG made in ComfyUI. ComfyUI writes the whole graph into the image metadata. This works if the metadata survived: many sites and messengers strip it on upload.
The models are whatever your template uses: SDXL, Flux and other checkpoints, your LoRAs, ControlNet, upscalers, image-to-video graphs. For video, our own production experience is WAN 2.2 image-to-video; more on that on the bulk video page.
No workflow yet, and you only need the result? Say so in the request. For video we can also work through commercial video models by API, without a template.
What ComfyUI workflow setup includes
Most of the work happens before the first image. A workflow file describes the graph, but not the environment it was built in. Setup means rebuilding that environment so the graph behaves the same way it did for its author.
- Graph review. We go through every node and model the workflow references and make a list: what is public, what is paid or closed, what is unknown.
- Custom nodes on pinned versions. Node packs are installed from git at fixed commits. Packs change often, and one update can rename an input or change a default, so a workflow that worked last month fails today. A pinned commit means the setup that passed the test stays the same through the whole bulk run.
- Models from where they actually live. Checkpoints, LoRAs, VAEs, upscalers: from Hugging Face, Civitai or a direct link you give us. A workflow refers to models by file name only, so we work out which exact file and version is meant before downloading.
- Unknown parts by hand first. A node pack nobody has heard of, or a model without a version page, gets examined by hand before it runs. Custom nodes are ordinary Python code with access to the machine, and we do not run them blind.
- Check on test frames. We render a few frames with your settings and compare them with the examples you sent. If they do not match, we find the cause (another model version, a node default, a missing LoRA) before going further.
The same process step by step, if you want to try it yourself first, is in the guide how to run a ComfyUI workflow from Civitai.
Closed models, paid models and licenses
Some things a workflow needs cannot simply be downloaded. A model may sit in paid early access on Civitai, be removed by its author, be available only through a paid API, or require you to accept a license first. Some workflows contain ComfyUI API nodes, which send every request to a paid external service. We find all of this during the graph review and tell you before the start, not after the invoice.
Licenses matter as much as files. Every model on Civitai carries its own permissions: whether generated images may be sold, whether the model may run on paid generation services, whether the author must be credited. Base models have their own licenses on top of that. If anything in the chain forbids commercial use, you hear about it before we start, and you decide: replace the model, ask the author for permission, or drop that part. Details are in the guide can you use Civitai models commercially.
The test batch and revisions
The test batch is a small run on your real inputs (your prompts, photos or references) at the final resolution and format. It answers two questions before you pay for volume: does the output look the way you want, and how long does one result take on our hardware.
You look at it and tell us what to change. We adjust the settings (sampler parameters, LoRA weights, resolution, prompt wording, upscaling) and show you a new test. Once you approve it, the settings are fixed. The bulk run uses exactly the approved setup, so the thousandth image comes from the same configuration as the test.
Running at volume on several GPUs
One graphics card is enough for a test; volume needs several working at once. Small and test orders run on our own GPUs. Large orders go to rented cloud GPUs: a fresh instance needs a few minutes to start, install the pinned nodes and download the models, and then several instances run in parallel, each with its own share of the inputs. Video models need much more memory than a typical image template, so the GPU type depends on your workflow.
We review the output by hand and drop the rejects: broken hands, melted faces, artifacts, frames that drift away from the approved style. The test batch you approved is the standard for the whole run.
Delivery is a ZIP archive or a link to a folder in cloud storage with a CDN. Output formats for social feeds and ads: 9:16 (1080×1920), 16:9, 1:1 and 4:5.
What this looks like for specific jobs:
- Bulk image generation on your checkpoint and LoRAs, from $0.10 per image.
- Bulk video generation: WAN 2.2 image-to-video, from $1.50 per clip. Every clip needs a start image.
- Character LoRA, when your character, mascot or product has to look the same in every frame.
How the work is split between GPUs and what breaks at scale is covered in the guide scaling a ComfyUI workflow to thousands of generations.
Cloud ComfyUI, a freelancer or GoraGen
Three usual ways to get volume out of a workflow:
| Cloud ComfyUI services | Freelancer | GoraGen | |
|---|---|---|---|
| Who sets up the workflow | You, within what the service supports | The freelancer | We do |
| Who runs the volume | You, job by job | Usually you, after handover | We do, on several GPUs in parallel |
| Who reviews the rejects | You | You | We do, by hand, before delivery |
| How the price works | GPU time by the hour or in credits, failed runs included | A one-off fee for the work | Per result; setup and test batch included |
| What you get | A running ComfyUI | A working setup | A reviewed batch of results |
| What you need to know | ComfyUI, VRAM limits, node debugging | Where it will run and who fixes it after updates | The result you want and the volume |
When the other options are better: if you want to experiment with the workflow yourself every day, a cloud ComfyUI service gives you that direct control. If you need a new custom node written, hire a ComfyUI developer: our job is the batch, not the graph. We fit when you know what result you want and the problem is getting it reliably, in volume, with the rejects already removed.
Prices and timing
| What | Price |
|---|---|
| Setup and test batch | included in the order |
| Images | from $0.10 per image |
| Video | from $1.50 per clip (5 seconds) |
| Character LoRA | from $99, one-time |
| Minimum order | $50 |
The exact price depends on how heavy the workflow is and on the volume: resolution, number of passes, upscaling and clip length all change the GPU time per result. Longer or higher-resolution video is quoted on request.
We do not quote timing blind. After the brief we know what the workflow needs; after the test batch we know how long one result takes. Then we agree on the price, payment and delivery date.
What to send
A complete brief saves a round of questions:
- The workflow: a link (Civitai, OpenArt, GitHub), a
.jsonfile or a PNG with metadata. - Examples of good output, ideally with seeds and settings, so we can compare the test frames.
- Private models and LoRAs: the files or access to them, if they are not public.
- Inputs: prompts, photos, references. For image-to-video, a start image for every clip.
- Volume and formats: how many results, which aspect ratios, one batch or a recurring order.
- What counts as a reject for you: anatomy, garbled text, off-brand colors.
- Where the results will be used, so we check the licenses against the real use.
- Your deadline, if you have one.
Send your workflow
Paste the workflow link and the volume into the request form. We review the graph and come back with what it needs, anything closed or restricted, and a ballpark price before the test batch.
