[BUG]: …not getting the ouput in response when using Save Image with extra metadata Node (crystools)

[Jhaaji]

Need help. How can I fetch output iamge or base64 url if the response is like:

{
"delayTime": 100,
"executionTime": 1571,
"id": "1c0cdba2-6102-4063-a845-a8174a825cac-e2",
"status": "COMPLETED",
"workerId": "1awbdsmhschpqw"
}

worker logs:

Prompt executed in 0.01 seconds
worker-comfyui - Status update: 0 items remaining in queue
worker-comfyui - Execution finished for prompt ec8d8f35-999c-443b-849a-9e2e4a0d2d01
worker-comfyui - Fetching history for prompt ec8d8f35-999c-443b-849a-9e2e4a0d2d01...
worker-comfyui - Processing 1 output nodes...
worker-comfyui - Node 101 contains 1 image(s)
worker-comfyui - Fetching image data: type=output, subfolder=, filename=ComfyUI.png_00001_.png
worker-comfyui - Successfully fetched image data for ComfyUI.png_00001_.png
worker-comfyui - Encoded ComfyUI.png_00001_.png as base64
worker-comfyui - WARNING: Node 101 produced unhandled output keys: ['text'].
worker-comfyui - --> If this output is useful, please consider opening an issue on GitHub to discuss adding support.
worker-comfyui - Closing websocket connection.
worker-comfyui - Job completed. Returning 1 image(s).
{"requestId": "1c0cdba2-6102-4063-a845-a8174a825cac-e2", "message": "Failed to return job results. | 400, message='Bad Request', url='https://api.runpod.ai/v2/bqjvbqagh2q90z/job-done/1awbdsmhschpqw/1c0cdba2-6102-4063-a845-a8174a825cac-e2?gpu=NVIDIA+GeForce+RTX+4090&isStream=false'", "level": "ERROR"}
{"requestId": "1c0cdba2-6102-4063-a845-a8174a825cac-e2", "message": "Finished.", "level": "INFO"}

[Jhaaji]

Update: not getting the ouptut with Save Image node as well

[Jhaaji]

worker-comfyui - Status update: 0 items remaining in queue
worker-comfyui - Execution finished for prompt a8a756b1-f322-4ef6-8979-28839e72b4b7
worker-comfyui - Fetching history for prompt a8a756b1-f322-4ef6-8979-28839e72b4b7...
worker-comfyui - Processing 1 output nodes...
worker-comfyui - Node 102 contains 1 image(s)
worker-comfyui - Fetching image data: type=output, subfolder=, filename=ComfyUIxyz_00001_.png
worker-comfyui - Successfully fetched image data for ComfyUIxyz_00001_.png
worker-comfyui - Encoded ComfyUIxyz_00001_.png as base64
worker-comfyui - Closing websocket connection.
worker-comfyui - Job completed. Returning 1 image(s).
{"requestId": "sync-6c6f0362-1843-49ad-ada0-b5db827c0be0-e2", "message": "Failed to return job results. | 400, message='Bad Request', url='https://api.runpod.ai/v2/bqjvbqagh2q90z/job-done/1awbdsmhschpqw/sync-6c6f0362-1843-49ad-ada0-b5db827c0be0-e2?gpu=NVIDIA+GeForce+RTX+4090&isStream=false'", "level": "ERROR"}
{"requestId": "sync-6c6f0362-1843-49ad-ada0-b5db827c0be0-e2", "message": "Finished.", "level": "INFO"}

[TimPietruskyRunPod]

Reproduced this on a fresh pod today (worker-comfyui 5.8.5-flux1-dev-fp8, RTX 4090, ComfyUI 0.17.2, ComfyUI-Crystools 1.27.4) and found two distinct findings — but I cannot reproduce the 400 Bad Request from a normal-sized image.

Finding 1 (separate bug, fix on the way)

Save image with extra metadata [Crystools] crashes with AttributeError: 'NoneType' object has no attribute 'copy' when with_workflow=True, because the worker doesn't forward extra_pnginfo in the /prompt payload. The fix lives in #225 (also unblocks #139). Without that fix, the crystools node only works with with_workflow=False.

But from your log (Job completed. Returning 1 image(s). came through fine), your crystools node clearly didn't crash, so your setup somehow had extra_pnginfo flowing through — either you're on an older worker version that included it, or with_workflow=false was set.

Finding 2 (your 400 — can't repro)

I ran your exact node (Save image with extra metadata [Crystools], filename_prefix="ComfyUI.png") inside the worker:

Test	with_workflow	response size	overhead	result
A	false	389,756 B (380 KB)	168 B	success
B (with handler patch from #225)	true	585,768 B (572 KB)	168 B	success

In both cases the response is well under 1 MB — nowhere near RunPod's 20 MB /runsync limit that would trigger a 400. The handler's output is just {images: [...]} with no workflow leakage. So whatever 400 you hit is not a "crystools always produces an oversized payload" bug.

What I need to actually fix this

Could you share:

1. Exact workflow JSON that triggered the 400. Even minimum-reproducer (no model loaders needed if you can substitute)
2. Image resolution and batch_size from the failing run (1024x1024, 2048x2048, batch >1, video output?)
3. worker-comfyui version you're running (5.x.y-base / 5.x.y-flux1-dev-fp8 / custom Dockerfile FROM?)
4. Any other custom nodes present in your image (ls /comfyui/custom_nodes/ from inside the container) — comfyui-easy-use, rgthree-comfy, etc. all change behavior

The 400 from job-done is RunPod's gateway rejecting the response payload. Reasons in order of likelihood: (a) base64'd PNG > 20 MB (large resolution or batch), (b) non-UTF8 bytes in the JSON, (c) something in the response that fails RunPod's validation. We need the workflow JSON to know which.