NODE-BASED WORKFLOW
Node-based AI workflows for image and video generation
A node-based AI workflow is a visual pipeline where each step — a text prompt, an uploaded image, a language model, an image or video model, or a utility such as cropping — is a node on a canvas. Connections pass one node’s output into the next node’s input, so a multi-step generation can be run, checked and reused.
Reviewed
How a node-based AI workflow works
- 01Add inputs
Put the brief in a Text node and upload product photos or clips to Image or Video input nodes.
- 02Connect model and tool nodes
Wire inputs into an LLM, image or video model node, then chain a result into the next node, for example a generated image as a video’s first frame. Check each model’s rate on the pricing page before a run; over the API, POST /workflows/{id}/estimate returns a per-node estimate.
- 03Run, review, reuse
Run the graph: results appear on each node, and task history records the credits actually charged. Save the workflow, swap the inputs and run it again, or call it through the API.
Nodes you can use today
This table follows the Add node panel in the RelayMe workspace today. Model names and prices come from the same live list as the pricing page at build time; 1,000 credits is about US$1.
| Type | Node | What it does | Price |
|---|---|---|---|
| Input | Text | A prompt or brief that feeds LLM, image or video nodes. | No credit charge |
| Input | Image | Upload JPG, PNG, WebP and other images (up to 30 MB each) as references or first frames. | No credit charge |
| Input | Video | Upload MP4, WebM or MOV files (up to 200 MB) as reference clips or a source for frame tools. | No credit charge |
| Input | Audio | Listed in the Add node panel but not enabled yet. | — |
| Model | LLM | Rewrites or expands a brief into prompts; the model selector lists the available text models. | Billed at the selected text model’s rate |
| Model | Image model | Nano Banana, Nano Banana Pro, Nano Banana 2, GPT Image 2.5, GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, GPT Image 2, Grok Imagine Image 2.0. Text-to-image and image-to-image with references; 1–4 candidates per run, each on its own output port; supported models add Upscale (2K / 4K) and Outpaint modes. | From 80 credits per image (standard); from 20 on the sale channel |
| Model | Image to prompt | Turns a connected image into an editable, structured image prompt with a vision text model. | Billed at the selected text model’s rate |
| Model | Video model | Grok Imagine, MiniMax-H3, Veo 3.1 Fast, Veo 3.1. Text-to-video and image-to-video; first-frame or reference inputs, duration, resolution and audio options depend on the model. | From 100 credits per second, by model, resolution and audio |
| Tool | Image crop | Select an area of an image and output it as a new image. | No credit charge |
| Tool | Video first / last frame | Extracts a video’s first or last frame as an image, for example to seed the next clip. | No credit charge |
| Output | (no separate node) | Each model or tool node shows its own result, can be downloaded, and exposes output ports for the next node; API runs return the final model nodes’ outputs. | — |
| Canvas | Label | A text note on the canvas; it is not executed. | No credit charge |
What it is, and what it is not
It is
- A hosted canvas: RelayMe calls third-party image, video and text models and does not train its own.
- Pay per generation in credits at each model’s listed rate; task history keeps the actual charge.
- Reusable: save a workflow, swap the inputs and run it again, or run it through the API.
It is not
- Not a general automation or coding tool: no custom code node, no loops, no custom nodes.
- Not a local runtime: no local checkpoints or LoRAs, and no offline use.
- Not every model: video is currently limited to Grok Imagine, MiniMax-H3, Veo 3.1 Fast, Veo 3.1.
- It cannot import workflow files from other tools, and there is no MCP server.
How this differs from automation tools like n8n
n8n also builds flows from nodes and connections, so the two are often compared. They solve different problems: n8n connects apps and services, while RelayMe is built for image and video generation. The n8n facts below come from its official pages, checked on 2026-10-05.
- n8n is a workflow automation and AI agent platform; its site lists over 500 integrations. source
- n8n Cloud bills by monthly workflow executions; one execution is one run of the whole workflow. source
- You can also self-host the Community Edition; the code is under the Sustainable Use License. source
- n8n does not run its own image or video models: its OpenAI and Google Gemini nodes offer “Generate an Image” and “Generate a Video” operations that call OpenAI or Google. source 1source 2
- These nodes normally use your own provider account and API key; some n8n Cloud plans also offer Gateway credits billed per request. source
When n8n fits better
- Generation is one step in a larger automation: triggers, CRMs, storage, approvals or scheduled batches.
- You already have OpenAI or Google accounts and want to pay them directly, or you need to self-host.
When RelayMe fits better
- The main job is making and iterating on images and video: results show on each node and models can be compared side by side.
- You want one credit balance across models from several vendors (video: Grok Imagine, MiniMax-H3, Veo 3.1 Fast, Veo 3.1) without managing API keys or servers.
Go further
Frequently asked questions
What is a node-based AI workflow?
Generation split into nodes on a canvas — inputs, language models, image or video models and tools — with connections that pass one node’s output into the next, so the whole chain can be run, checked and reused.
Which nodes does RelayMe have today?
Inputs: Text, Image and Video (Audio is not enabled yet). Models: LLM, image models (Nano Banana, Nano Banana Pro, Nano Banana 2, GPT Image 2.5, GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, GPT Image 2, Grok Imagine Image 2.0, with Upscale and Outpaint modes), Image to prompt, and video models (Grok Imagine, MiniMax-H3, Veo 3.1 Fast, Veo 3.1). Tools: Image crop and video first / last frame. Plus canvas labels.
How much does a node-based workflow cost to run?
Only generation nodes spend credits: each image, video or LLM step at its model’s rate (video from 100 credits per second; images from 80 credits at the standard rate or 20 on the sale channel). Inputs, tools, labels and saving a workflow do not. 1,000 credits is about US$1.
Can I import a ComfyUI or other workflow file?
No. RelayMe cannot open workflow files from other tools; rebuild the graph with the nodes above. Steps that rely on custom nodes or code have no equivalent.
Is RelayMe the same kind of tool as n8n?
No. n8n is a general workflow automation platform billed per workflow execution; for images or video it calls services such as OpenAI or Google. RelayMe only does image and video generation: models from several vendors on one canvas, billed per generation, but with no triggers, schedules or third-party app integrations. Use n8n when generation is one step in a bigger process, and RelayMe when producing the images and video is the main job.
Can I run a workflow from code?
Yes. Once saved, a workflow can be run over the HTTP API: read its input schema, estimate the cost, submit a run and poll for the result. Browser-side tool nodes such as Image crop and first / last frame cannot run on the server and are listed in unsupportedNodes.
Build your first workflow
Start with one Text node and one image model node, check the result and the credits, then add a video node.