Promo image

What this is

A text-to-image graph on FLUX.1-dev in fp8. You give it a sentence; it gives you one picture, sized for a tool page. Optionally it takes one of your screenshots as a reference and works image-to-image from it instead.

It is the cheapest thing here to run, and the best one to try first: about a cent and a half of GPU on a cold worker.

Run it on your own machine

You need ComfyUI and one file.

• flux1-dev-fp8.safetensors → models/checkpoints/ — about 17 GB

From Comfy-Org/flux1-dev — no Hugging Face token needed. A 12 GB card runs it; 8 GB works with --lowvram and patience.

Then download the graph below, Workflow → Open it in ComfyUI, and replace the {{placeholders}} with real values — {{prompt}} with your sentence, {{seed}} with a number, {{width}}/{{height}} with a size. That is the whole difference between our copy and yours: we fill those in per job.

Run it on RunPod

No models to download at all — RunPod publishes a ComfyUI worker with this exact checkpoint baked into the image.

1. Sign up at RunPod and add a little credit.
2. Serverless → New Endpoint → Import from Docker Registry, image runpod/worker-comfyui:5.10.0-flux1-dev-fp8.
3. Container disk 40 GB (the weights are in the image), GPU any 24 GB card, max workers 1, idle timeout 5 s.
4. POST https://api.runpod.ai/v2/<endpoint>/run with {"input": {"workflow": <the graph below>}}.

Set the idle timeout low. A flex worker bills from the moment it starts until it shuts down, so at one job every few minutes a 60-second idle tail costs more than the work does. Ours sat at $1.11/hour with nothing running before we noticed.

What it costs

About $0.006 warm, $0.014 cold, measured against the RunPod balance rather than worked out from a rate card.