Debugging strategies

[dcmaf]

Any advise on how to debug? I can get images to inject using the default workflow and it looks promising, but haven't been able to get it to work with a custom workflow (with placeholders as discussed in the docs). Even the default workflow can be hit or miss depending on the system prompt (I'm using an Author's Note). Even the same prompt in different chats sometimes gets image tags generated and sometimes doesn't.

Getting the AI (I'm using Mistral-Nemo-12B) to generate the image tag in the proper format is tricky - I can see what is coming back from the AI in the logs, but what ComfyInject does after that is a black box and I don't see any logs to help, just an [Image generation failed] message in the chat.

I have checked the image gallery and the metadata on the images that were successfully created when I open them. There I see AR, SHOT, and Filename all as N/A.

Also, is the extension dependent on the description being tags? I am attempting to use Flux.2 in my custom workflow, and it works with natural language prompts.