ComfyUI Image quantity / batch size issue when sending request to serverless endpoint
I'm not able to generate multiple images from a prompt / request to the endpoint using a ComfyUI workflow. We have added a variable for the ""batch_size": " value in our workflow, but it only seems to generate one image regardless of the batch_size we give it.
This is our Github repo for the runpod worker:
https://github.com/sozanski1988/runpod-worker-comfyui/
And the workflow json is here:
https://github.com/sozanski1988/runpod-worker-comfyui/blob/main/test_resources/workflows/default_workflow.json
Is there a special way that we need to alter the workflow to be able to generate urls for multiple image outputs?
P.S. I've also added an example request and serverless endpoint logs for reference.
11 Replies
Upload all of the results
Sorry, I'm unsure what you mean. Which results do you want me to upload?
The image results
To s3, and then return the url
Isn't it how it should be working?
Does it return the specified amount inside the comfyui? Or it's because of the python handler for runpod serverless?
In the request logs it just seems to be outputting one image result, even when batch_size should be set to a higher quantity
I'm not sure this is relevanr, Im unable to run your comfyui setup, are you Able to debug the comfyui running inside your worker? Or run it locally to see if the work flow actually saves more than 1 image
If yes, then it's an issue in your handler code, if not, then fix your work flow to save more than 1 files
Yes, the ComfyUI workflow works locally and is outputting multiple images
Okay. Check the handler code where it uploads the image results that are saved
This is the handler code:
https://github.com/sozanski1988/runpod-worker-comfyui/blob/main/src/rp_handler.py
When I send a request on Runpod it only seems to save the output of the last image generated:
2025-01-08T13:10:36.011632514Z Prompt executed in 15.89 seconds
2025-01-08T13:10:36.128850300Z runpod-worker-comfy - image generation is done
2025-01-08T13:10:36.128895408Z runpod-worker-comfy - /comfyui/output/ComfyUI00012.png
2025-01-08T13:10:36.133662502Z runpod-worker-comfy - the image was generated and converted to base64
2025-01-08T13:10:37.512337892Z {"requestId": "89b2d639-6fee-4aab-a1bc-6ea766b85cd9-e1", "message": "Finished.", "level": "INFO"}
(I'm using the handler file you provided earlier btw, the "blib_la" one
(I set the batch size to 12 images here)
We get the image output as a base64 encoded string (we didnt set up AWS as we use Firestore and we thinking to store the image file their after we've retrieved it from the Runpod worker)