[Bug]: Often (most of the time) will not generate image

[doc3d]

Pre-Issue Checklist

• I am using SillyTavern 1.16.0 or newer
• This issue is NOT caused by a custom workflow
• I read the README and FAQ

SillyTavern Version

1.16.0

ComfyUI Host

127:0.0.1:8188

Describe the bug

Ive been using this (and I really like it) but for some reason it works very randomly. I get the [Generating image 1/1...] but no image is produced. I can see in ComfyUI console that it nothing is sent to it. This can happen on several messages in a row, then suddenly it generates one or two times before going back to not displaying. I would say i get 1 image for every 10-15 requests (it is set to produce one image per message)
I am using GLM 4.6 via OpenRouter

The idea is greeat, I just wish it was a bit more stable. Perhaps also a way to manually trigger or have it post images less frequently. In every image (if it works) seems a bit redundant and if I try to tell it to include images "when narratively appropriate" it still tries to generate a new image in every message.

Also wondering if its possible to get it working remotely? I have ngrok active so I can use my phone or any external device to connect to my ST instance running at home, but this always return a "image generation failed" in every message, regardless of device.

Steps to reproduce

whenever a message is sent, I see the [Generating image 1/1...]
If I look in ComfyUI console nothing is sent
No image is rendered and I can start typing the next message

Browser Console Errors

It keeps showing the following error: 

[ComfyInject] Marker has 3 segment(s), expected 4. Skipping.

Screenshots

No response

[doc3d]

update. I might have gotten it working again. Not sure what the issue was. I tried redownloading it, and replacing the files manually. After doing that, it seems to work on every iteration again. I didnt change any files (to my knowledge) so I have no idea, but hey, thats computers for you.. I will leave the issue open, in case you have any feedback as to what might have happened or have any comments on the other pprats I mentioned. Thank you for a truly great addition to ST!

...aaaand I spoke too soon. Just as I was writing this comment, the issue returned (The F12 console goes back to [ComfyInject] Marker has 3 segment(s), expected 4. Skipping.

After a bit more testing, it seems to behave again.. Its very weird, because I am not changing anything, it just suddenly stops or starts working on a per message basis, but its a lot better now than when I started this post..

[doc3d]

It has stopped working again, I havent recieved an image in the last 15-20 images.. the error in the F12 console is:

[ComfyInject] Processing message 954
eventemitter.js:149 TypeError: Cannot read properties of null (reading 'seed')
at getLastSavedSeed (state.js:47:46)
at resolveSeed (state.js:72:31)
at parseMarkerContent (parse.js:66:18)
at processAllImageMarkers (parse.js:90:24)
at processMessage (dom.js:133:27)
at EventEmitter. (dom.js:313:15)
at EventEmitter.emit (eventemitter.js:146:36)
at async StreamingProcessor.onFinishStreaming (script.js:3663:13)
at async finishGenerating (script.js:5269:17)
at async sendTextareaMessage (script.js:1693:22)
EventEmitter.emit @ eventemitter.js:149
await in EventEmitter.emit
onFinishStreaming @ script.js:3663
await in onFinishStreaming
finishGenerating @ script.js:5269Understand this error
eventemitter.js:150 Error in event listener

[myonmu0]

They are working on this but I make a mitigation to try it now.
Edit the function parseMarkerContent on the ComfyInject/src/parse.js to:

function parseMarkerContent(innerContent, messageIndex) {

    const rawPrompt = innerContent;

    // Validate prompt — if empty we really can't do anything useful
    if (!rawPrompt) {
        console.warn("[ComfyInject] Marker has an empty prompt. Skipping.");
        return null;
    }

    const ar = DEFAULT_AR;
    const shot = DEFAULT_SHOT;
    const seed = resolveSeed("RANDOM".toUpperCase(), messageIndex);

    return { prompt: rawPrompt, ar, shot, seed };
}

On the SillyTavern chat, make LLM write as follow format: [[IMG: PROMPT ]]

Note:

• If you don't want SHOT prompt added in your PROMPT, ComfyInject > Advanced Settings > Shot Tags: Lock Shot to WIDE and delete WIDE content.
• Extension-PromptInspector may be useful to see the IMG tag that LLM wrote, for some reason my GLM are adding something like "| 12345" on the end of the prompt.

[doc3d]

OK, just so Im clear. I edited the parse.js file. I took out the paragraph that was there and pasted what you wrote above instead. Then I edited the System Prompt from

The marker must follow this exact format:
[[IMG: PROMPT | AR | SHOT | SEED ]]

to:

The marker must follow this exact format:
[[IMG: PROMPT ]]

I then tried to write another message. The F12 console gives me the following error:

[ComfyInject] Processing message 1159
TypeError: Cannot read properties of null (reading 'seed')
at getLastSavedSeed (state.js:47:46)
at resolveSeed (state.js:72:31)
at parseMarkerContent (parse.js:66:18)
at processAllImageMarkers (parse.js:90:24)
at processMessage (dom.js:133:27)
at EventEmitter. (dom.js:313:15)
at EventEmitter.emit (eventemitter.js:146:36)
at async StreamingProcessor.onFinishStreaming (script.js:3663:13)
at async finishGenerating (script.js:5269:17)
at async sendTextareaMessage (script.js:1693:22)

No image and nothing is sent to ComfyUI, that console is clean and empty..

[myonmu0]

After editing ComfyInject/src/parse.js, F5 or restart SillyTavern?

Since the error at parseMarkerContent (parse.js:66:18) is identical before and after editing, may be parse.js arent reloaded or not edited?

For me parse.js are on SillyTavern/data/default-user/extensions/ComfyInject/src/parse.js but this may vary.

[doc3d]

After updating and restarting ComfyUI, and doing the same with ST, I get the following error: I did get it working yesterday, but had to reboot my computer today, so it just seems very random. It can work for a while, then suddenly it stops working.. and if I just keep chatting, then suddenly it can start generating again anfter like a day.. its very strange and I have no idea how to document it, since there is no change. I just write as I usually do, it either works or it doesnt, no apprant logic that I can see.

[ComfyInject] Image generation failed for marker 1: Error: [ComfyInject] Timed out waiting for image after 60 attempts
at pollForResult (comfy.js:109:11)
at async generateImage (comfy.js:182:37)
at async processAllImageMarkers (parse.js:78:28)
at async processMessage (dom.js:133:21)
at async EventEmitter. (dom.js:313:9)
at async EventEmitter.emit (eventemitter.js:146:17)
at async StreamingProcessor.onFinishStreaming (script.js:3663:13)
at async finishGenerating (script.js:5269:17)
at async SimpleMutex.sendTextareaMessage [as callback] (script.js:1693:22)
at async SimpleMutex.update (SimpleMutex.js:38:13)

[myonmu0]

This error are telling that ComfyUI image generation don't completed in 60 second for some reason.

You can increase the timeout:
ComfyInject > Advanced Settings > Max Poll Attempts (edit this, default is 180 second)

[Spadic21]

Thanks for all the testing, it helps a lot.

There are a few different things going on in this issue:

• The Marker has 3 segment(s), expected 4 message means your LLM isn't always outputting the full marker format, so those get skipped. That's why it feels random, it just depends on what the model decides to output for each message.
• The Cannot read properties of null (reading 'seed') error is a separate issue on my side. That shouldn't happen even when a marker gets skipped, so I'll be looking into that.
• The timeout error as explained by @myonmu0 is just that the generation was started, but ComfyUI didn't return a finished image in time.

I'm working on improving the parser so it handles incomplete markers better instead of just skipping or breaking like this.