Creator OS
ai-image-generationcomfyuicontent-automationminimaxsocial-media-workflow

MiniMax H3-Image ComfyUI Workflow: 4K Images in 10 Seconds

MiniMax H3-Image makes 4K images in about 10 seconds, and the ComfyUI workflow drops them into automated social pipelines. Here is the setup and how to batch it.

Creator OS · November 12, 2026 · 8 min read

A ComfyUI image workflow for social media is a node graph that generates images from a prompt list, upscales them, and hands finished files to your scheduling pipeline. Build it once, then batch visuals for posts, thumbnails and ads without opening a design app for every asset.

prompt list ──► ComfyUI graph ──► PNG (local folder)
                    │
                    ├─ upscale node
                    ├─ caption / alt text node
                    └─ save + webhook ──► Creator OS media id
                                              │
                        posts:create ──────────┘
                        (11 platforms + Skool)
One graph, a batch of images, one hand-off into scheduling.

Why Generate Post Images Locally at All

The point of a local graph is volume under your own control: you can produce a fresh image per post instead of reusing the same three files.

Packed as a ComfyUI workflow, image generation becomes a repeatable pipeline step rather than a one-off experiment. That matters for social because visuals are the slow part of the day for most creators and small agencies: someone has to find a stock photo, crop it four ways, and export at the right size.

Running the graph on your own hardware also means the files never leave your machine until you choose to upload them. You can generate a month of covers in one sitting, on your own GPU, at whatever speed your setup gives you. Time a single run before you plan a batch, because GPU, model quantisation and the upscale step all move that number.

Get started for $0

How a ComfyUI Image Workflow Works

A working graph is four blocks: prompt source, sampler, upscaler, output.

  1. Prompt source. Either a single text node you edit by hand, or a loader that reads a JSON or CSV list of product names, hooks and angles. The loader is what makes batching possible.
  2. Sampler and model nodes. Your checkpoint loads here with your steps and guidance values. Keep these in one group so you can swap the model later without rewiring the output side.
  3. Upscale pass. Base generation, then an upscale node. Doing the upscale as a separate node keeps the base render fast and lets you change target size per platform without touching the sampler.
  4. Output. A save node writes to a dated folder, and a filename node builds predictable names like ig-reel-cover-001.png. Predictable names matter later: your script uploads by filename pattern, not by sorting through a folder by eye.

Once the graph is stable, ComfyUI can run it headless from a script, which is the bridge to automation. The graph does not need to know anything about social platforms. It just needs to drop finished files somewhere your posting pipeline can read.

How to Install a Model and Workflow in ComfyUI Step by Step

Install ComfyUI first, then add the node pack and checkpoint for the model you picked before you try to open anyone’s shared workflow file.

  1. Set up ComfyUI. Install ComfyUI and confirm the example graph runs on your machine. If the default graphs fail, fix the Python and CUDA environment before adding a second model.
  2. Install the node pack. Add the node pack for your chosen model with ComfyUI Manager, or clone it into the custom_nodes folder and install its requirements.
  3. Place the checkpoint. Download the model files and put them in the folders the node pack expects. Most load errors are a file in the wrong directory, not a broken model.
  4. Restart and reload. Restart ComfyUI and check the console for missing dependency warnings. Reload the page after the restart so the browser picks up new nodes.
  5. Import the workflow JSON. Drag the workflow file onto the canvas. If any node shows red, right click it and read the missing type name: that tells you which pack is absent.
  6. Run one low resolution image. Generate a single small image and confirm it looks correct. Only add the upscale pass once the base render is clean.
  7. Test the upscale. Raise the target size and time the run. If memory runs out, lower the batch size to one image per pass before changing anything else.

Write the working steps down in a short doc. When you share the graph with a teammate, a five line install note saves an afternoon.

Can You Batch Images for Social Media Without Manual Design

Yes, if the prompts are data and the outputs are named by platform.

The pattern is a prompt table with one row per post. Columns hold the product or topic, the hook, the aspect ratio, and the target platform. The loader feeds rows into the graph one at a time, the save node writes a platform-sized file, and the whole run finishes without a person clicking generate.

Platform target Shape to render Why it matters
Instagram Reel cover Vertical 9:16 Reels and Stories are vertical; a wide crop wastes the frame
Instagram feed or carousel Square or 4:5 Feed slots read better with one clear subject
YouTube thumbnail 16:9 Custom thumbnails work on long form YouTube uploads
TikTok video cover Vertical 9:16 TikTok accepts a custom cover image with the upload
Blog hero Wide 16:9 WordPress and Shopify article images

Two practical rules keep batches clean. First, render every image one step larger than the platform needs, then let the upload crop rather than the model. Second, keep the prompt table in version control so you can rerun a month and reproduce it.

How to Connect ComfyUI Output to Your Social Media Workflow

The hand-off is a file upload and a post call, and both are scriptable with the Creator OS CLI.

When ComfyUI finishes a run, upload the PNG once with creatoros media:upload. That returns a media id you can reuse across platforms, so one render serves Instagram, TikTok, X and LinkedIn without uploading four times.

# 1. Upload the generated image once
creatoros media:upload ./out/ig-reel-cover-001.png

# 2. Check the caption against platform rules
creatoros validate:post-length --text "New drop: the winter restock is live."

# 3. Create the post on the platforms you want
creatoros posts:create \
  --text "New drop: the winter restock is live." \
  --platforms instagram,tiktok,x \
  --media med_12345 \
  --scheduledAt 2026-03-04T14:00:00Z --timezone America/Toronto

If you would rather have an AI agent run the loop, Creator OS runs a hosted MCP server at https://mcp.creatoros.ca/mcp, so a Claude or Cursor session can call upload_media_from_url and create_post directly. The docs at the MCP page cover the connection steps, and the CLI reference lives at the CLI docs.

For a full agent setup, the open source Social Agents repo runs your accounts on Creator OS: read the Social Agents walkthrough. Kevin walks through builds like this on the KevBuildsApps YouTube channel, and the launch video is at youtube.com/watch?v=-QBJH_PK3pY.

What a Social Media Approval Workflow Looks Like Around AI Images

Fast generation creates a new problem: someone still has to approve what goes out.

A simple flow has three states. Generated images land in a review folder. A person approves or rejects each file, and the approved list feeds the upload step. Rejected files never reach the posting call, so nothing ships half finished.

In Creator OS you can create posts as drafts with --draft, then list them for review:

creatoros posts:list --status draft --limit 20

An agency can keep the drafting and the approval in different hands. Agency plans include a scoped API key per client workspace, so a contractor generating images for one client cannot publish to another client’s accounts. The posting side of approval, including how to structure review steps across a team, is covered in the agency approval workflow post.

If your pipeline mixes generated stills with video covers and edited clips, the open source OPEN VIDEO EDIT skills turn a raw talking head clip into an animated short locally with Python and ffmpeg. It pairs well with a ComfyUI graph: one produces stills, the other produces motion.

ComfyUI Image Workflow FAQ

What is the best ComfyUI image workflow for social media posts?

The best one is boring: a prompt loader, the sampler, a separate upscale node, and a save node with predictable filenames. That structure lets you batch a prompt table and reuse one media id across platforms. Keep the sampler and upscale steps in separate node groups so you can change target size without rewiring.

Can ComfyUI run locally or does it need an API?

ComfyUI runs graphs on your own hardware, and community node packs wrap models so they load as local nodes. Whether a specific model runs on your machine depends on your GPU memory and the checkpoint you download. Test a single low resolution image before you attempt a large batch.

How do I batch generate images in ComfyUI for multiple platforms?

Put your prompts and target shapes in a CSV or JSON table, load it into the graph, and let a filename node tag each output by platform. Run the batch once, then upload the finished files with creatoros media:upload and reuse the returned media id across platforms. Render slightly larger than each platform needs and let the upload crop.

How do I add a social media approval workflow to AI-generated images?

Create posts as drafts with creatoros posts:create --draft and list them with creatoros posts:list --status draft before anything publishes. On agency plans, a scoped API key per client workspace keeps each client’s queue separate. A reviewer then schedules or rejects each draft.

Is a ComfyUI workflow better than n8n for AI image posting to social?

They do different jobs. ComfyUI is an image generation graph; n8n is a general automation platform. Creator OS ships a hosted MCP server and a CLI, so an AI agent can generate a caption, upload the image and schedule the post in one session, and it connects to the social accounts directly rather than through a separate automation layer. See Creator OS vs n8n for social media workflows for the detailed split.

Get started with Creator OS for automated social pipelines

Once your ComfyUI graph outputs finished images, the remaining work is upload, caption, schedule and reply. Creator OS handles that side: 11 platforms plus Skool from one place, usable from the web app, the iOS app, a REST API, the CLI, a Claude Code plugin and a hosted MCP server.

The API, the MCP server and the CLI are included on every plan. Creator plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. Agency and YouTube network plans give each client or channel its own workspace and scoped API key. Start with a free trial at https://www.creatoros.ca/sign-up.

Get started for $0
Free trial for new subscribers. Cancel anytime.

Keep reading