[BUG]: Unable to get more than 1 output image from a workflow

[julien-moment]

Describe the bug

I have a workflow with one single Save image node (core) at the end.
This workflow is supposed to generate 4 images using Batch-condition Batch-String custom node to send 4 different prompts.
It works properly in the Web UI generating all 4 images but when used with the runpod-worker-comfy it only sends 1 image in the result message (Base64 encoded) instead of 4, but I can see that the 4 imgs have been generated in the output folder.
So it seems that the worker API isn't able to get the 3 other images... ?

I have the same issue running everything on runpod and on my local server.

Repro MVP (Minimal Viable Procedure)

Here is a simple workflow example, generating 4 images using Batch Condition node :
Batch Image Generation.json

Expected behavior

I expect the result message having all 4 images base64 encoded.

Screenshots

Versions (please complete the following information):

• ComfyUI version: 0.3.12
• Host OS: Ubuntu 22.04
• Docker version: ??

Additional context

[bluematter]

Same issue, the problem is here

output_images = {}

for node_id, node_output in outputs.items():
    if "images" in node_output:
        for image in node_output["images"]:
            output_images = os.path.join(image["subfolder"], image["filename"])

Every time the loop iterates over an image, it assigns a new value to output_images instead of adding to a list. Thus, if you have multiple images, only the final image’s path will remain in output_images.

The proposed solution from chatGPT is

def process_output_images(outputs, job_id):
    COMFY_OUTPUT_PATH = os.environ.get("COMFY_OUTPUT_PATH", "/comfyui/output")
    
    # Use a list to accumulate image paths
    output_images = []

    for node_id, node_output in outputs.items():
        if "images" in node_output:
            for image in node_output["images"]:
                image_path = os.path.join(image["subfolder"], image["filename"])
                output_images.append(image_path)

    print("runpod-worker-comfy - image generation is done")
    
    # Check if any images were found
    if not output_images:
        print("runpod-worker-comfy - no images found in the outputs")
        return {
            "status": "error",
            "message": "No images were generated in the output.",
        }

    # Process each image: upload or encode
    processed_images = []
    for rel_path in output_images:
        local_image_path = os.path.join(COMFY_OUTPUT_PATH, rel_path)
        print(f"runpod-worker-comfy - Processing image at {local_image_path}")
    
        if os.path.exists(local_image_path):
            if os.environ.get("BUCKET_ENDPOINT_URL", False):
                image_result = rp_upload.upload_image(job_id, local_image_path)
                print("runpod-worker-comfy - the image was generated and uploaded to AWS S3")
            else:
                image_result = base64_encode(local_image_path)
                print("runpod-worker-comfy - the image was generated and converted to base64")
            processed_images.append(image_result)
        else:
            print(f"runpod-worker-comfy - the image does not exist: {local_image_path}")
            processed_images.append(f"Error: Image does not exist at {local_image_path}")

    return {
        "status": "success",
        "message": processed_images,
    }

Im forking and will submit a PR if it fixes it

[bluematter]

Submitted a PR works for me #96 enjoy!

[TimPietrusky]

@julien-moment @bluematter sorry for the delay here!

the problem will be fixed with #118 as we then return all the images that were generated, not just one

[TimPietrusky]

🎉 This issue has been resolved in version 5.0.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀