Pose transfer from a reference photo with upscaling

Pose transfer from a reference photo with upscaling

An n8n workflow with an upload form for two pictures: a photo of a person and a picture of a pose. A RunningHub AI app
redraws the person in that pose, and a second app enlarges the result until its shorter side is 1440 pixels. Every step
runs on RunningHub, through its API.

What is inside

• workflow.json: the workflow as n8n exports it, 22 nodes and no sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The key's placeholder in API Setup was YOUR_RUNNINGHUB_API. It is now YOUR_RUNNINGHUB_API_KEY, so it says what
  goes there. Nothing else changed.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A RunningHub account with a balance, and its API key. Both apps run as paid tasks there.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Open API Setup and replace YOUR_RUNNINGHUB_API_KEY with your RunningHub key. Every RunningHub call reads it from
   there, so it is the only place to change.
3. Save, press Execute workflow and upload two pictures in the test form that opens.
4. Publish the workflow when you want the form's lasting address.

How a run goes

1. On form submission takes two required files, Upload Original Image and Upload Pose Image, as JPG, PNG or WebP.
2. getBinaryFile passes both files on in one item. uploadOriginalCharacter and uploadPostureDiagram send them to
   RunningHub's /task/openapi/upload side by side, and each gets back a stored fileName.
3. Merge and Aggregate wait for both uploads. createTask then starts the RunningHub AI app 1986280816420552706 at
   /task/openapi/ai-app/run. Its node 23 gets the person, node 41 the pose and node 34 the instruction, which asks for the
   person in the first picture to take the pose in the second.
4. Wait pauses a minute and a half before checkStatus asks for the task's state. Switch sends QUEUED and RUNNING
   to Wait1, which checks again every minute, and SUCCESS on to getOutput.
5. getOutput reads the result's address, downloadImage fetches the picture, and uploadNewImage sends it back to
   RunningHub as a new file.
6. createUpscaleTask starts the second app, 1961632883674955778, with that file in node 3 and 1440 in node 15, the
   length of the shorter side.
7. Wait2 pauses a minute, checkStatus1 asks again, and Switch1 loops through Wait3 every 30 seconds until the task
   is done.
8. getOutput1 reads the address, and downloadFinalImage fetches the finished picture as a file in n8n.

Change the look

• The instruction: the text value in createTask's JSON body. Say more about the clothes, the background or the
  framing if the result drifts.
• The final size: 1440 in createUpscaleTask, the shorter side in pixels.
• The apps: webappId in createTask and createUpscaleTask. The nodeId numbers belong to each app, so check them on
  the app's page in RunningHub when you swap one.
• How long it waits: Wait (90 seconds), Wait1 and Wait2 (one minute each) and Wait3 (30 seconds).

Good to know

• The picture stays in n8n. No node shows it on the form or saves it anywhere, so open downloadFinalImage under
  Executions to download it. To keep each one, add a Google Drive, Dropbox or email node after downloadFinalImage.
• A failed task stops the run quietly: Switch and Switch1 have nothing on their first output, FAILED. Any state
  they do not test ends the run with an error.
• RunningHub bills both tasks to your account, the pose change and the enlargement.
• Once the workflow is published, anyone with the form's address can start a run on your RunningHub balance.
• Both pictures go to RunningHub, and so does the result before it is enlarged. Use only photos you have the right to
  use, and ask the people in them first.
• The RunningHub key sits in API Setup and travels in each request's body, so take it out before you share the
  workflow.