Adetailer + controlnet generates noise after first image in a batch

[rpillala]

Hello! I love controlnet. It's such a great extension and has improved my generations 100x. In fact, without it I don't know if I would still be using stable diffusion at all. I said the same thing about afterdetailer at their github but it's true.

I tried searching for this error and didn't find anything. I'm also not sure if the problem is in controlnet or after detailer. I have posted this same messsage on the adetailer github.

Using my local A111 installation, when I use a single control image to generate batch count > 1, the first image generates normally, and every image after that one is noise. This only happens if I use controlnet and adetailer together. Has anyone else had this issue? I have tried a1111 1.9.0 and 1.9.3, and I have tried the "passthrough" controlnet option in adetailer, and the stylealign option in controlnet. With stylealign active, images after the first one do not use controlnet at all, but they aren't noise. I'm using ControlNet v1.1.448 and ADetailer v24.4.2. And, if it matters, I am using python: 3.10.8  • torch: 2.1.2+cu121 . And I've tried with and without dynamic prompts active, but there are no wildcards in the prompt "a man in a business suit." I'm not sure how to downgrade controlnet or adetailer to try a previous version. I was able to successfully generate batch count = 2 at Thinkdiffusion, where they are using earlier versions of the extensions.

I don't see any error messages in the console output:

INFO:sd_dynamic_prompts.dynamic_prompting:Prompt` matrix will create 2 images in a total of 2 batches.

2024-05-14 08:25:40,186 - ControlNet - INFO - unit_separate = False, style_align = False
2024-05-14 08:25:40,586 - ControlNet - INFO - Loading model: control_v11p_sd15_openpose [cab727d4]
2024-05-14 08:25:41,662 - ControlNet - INFO - Loaded state_dict from [F:\stable-diffusion-webui\models\ControlNet\control_v11p_sd15_openpose.pth]
2024-05-14 08:25:41,663 - ControlNet - INFO - controlnet_default_config
2024-05-14 08:25:45,466 - ControlNet - INFO - ControlNet model control_v11p_sd15_openpose cab727d4 loaded.
2024-05-14 08:25:45,482 - ControlNet - INFO - Using preprocessor: openpose_full
2024-05-14 08:25:45,483 - ControlNet - INFO - preprocessor resolution = 512
2024-05-14 08:25:49,258 - ControlNet - INFO - ControlNet Hooked - Time = 9.07895040512085
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [00:13<00:00, 1.89it/s]

0: 640x448 1 face, 313.9ms
Speed: 6.0ms preprocess, 313.9ms inference, 40.9ms postprocess per image at shape (1, 3, 640, 448)
100%|██████████████████████████████████████████████████████████████████████████████████| 11/11 [00:03<00:00, 3.04it/s]
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
mediapipe: 1 detected.
100%|██████████████████████████████████████████████████████████████████████████████████| 11/11 [00:03<00:00, 2.97it/s]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 2 images in a total of 2 batches.
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [00:06<00:00, 3.65it/s]

0: 640x448 (no detections), 15.9ms
Speed: 3.0ms preprocess, 15.9ms inference, 2.0ms postprocess per image at shape (1, 3, 640, 448)
[-] ADetailer: nothing detected on image 2 with 1st settings.
[-] ADetailer: nothing detected on image 2 with 2nd settings.
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 2 images in a total of 2 batches.