Black image issue when using IP Adapter | Invoke | Page 1

young mulch Jan 7, 2024, 3:04 PM

#

I'm getting black images every time I use an IP Adapter; it works fine otherwise. It happens both with SD15 and SDXL, and I downloaded new IP Adapter models from InvokeAI's Huggingface repo and it does the black screen with all of them. (And yes, I'm using FP32). Using an RTX 3060. Any ideas?

viral radishBOT Jan 7, 2024, 3:04 PM

#

Hi @young mulch. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:

#

Based on the information provided in the extracts, it seems that this issue of black images has been reported by other users as well. Here are a few suggestions that might help:

Change the Precision Setting: Some users have reported that changing the precision setting to 'float32' has resolved the issue. However, you mentioned that you're already using 'float32'. It might be worth trying 'fp16' or 'auto' to see if that makes a difference.
Update InvokeAI: Make sure you're using the latest version of InvokeAI. Some users have reported that updating to the latest version resolved their issues.
Reinstall IP Adapter Models: There have been instances where the issue was resolved by deleting and redownloading the IP Adapter models. Make sure you're downloading the correct models for your base model (SD15 or SDXL). You can find the models on the InvokeAI Huggingface repo (https://huggingface.co/InvokeAI).
Check for Errors: Look for any error messages or warnings in the console output when you run the model. This could provide clues as to what's going wrong.

If none of these suggestions work, I recommend reporting the issue on the InvokeAI GitHub Issues page (https://github.com/invoke-ai/InvokeAI/issues). Include as much detail as possible, including your OS, GPU, VRAM, the version of InvokeAI you're using, and any error messages you're seeing. This will help the developers diagnose and fix the issue.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@young mulch kapa.ai is still learning and improving, please let us know how it did by reacting below

young mulch Jan 7, 2024, 3:39 PM

#

Changing the precision to FP16 worked with SDXL with sdxl-vae-fp16 fix. It failed on SD15 with the default VAE. @viral radish

viral radishBOT Jan 7, 2024, 3:39 PM

#

Let me check the knowledge base for you. One moment! :books:

#

Based on the information from the extracts, it seems that the SDXL model works with 'fp16' precision when used with the 'sdxl-vae-fp16-fix' VAE. However, the SD15 model fails with the default VAE when 'fp16' precision is used.

This could be due to the fact that the default VAEs that come with the models are not compatible with 'fp16' precision. As mentioned in the GitHub issue #4831 (https://github.com/invoke-ai/InvokeAI/issues/4831), you need to use a custom VAE, like 'sdxl-vae-fp16-fix', to use 'fp16' precision.

If you want to use 'fp16' precision with the SD15 model, you might need to find or create a custom VAE that is compatible with 'fp16' precision, similar to the 'sdxl-vae-fp16-fix' VAE for the SDXL model.

Alternatively, you could try using 'fp32' precision with the SD15 model and the default VAE, as this has been reported to work in some cases, as seen in GitHub issue #4083 (https://github.com/invoke-ai/InvokeAI/issues/4083).

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@young mulch kapa.ai is still learning and improving, please let us know how it did by reacting below

young mulch Jan 7, 2024, 3:42 PM

#

If you refer back to my original post, it was failing on SD15 using fp32 with the default VAE. @viral radish

viral radishBOT Jan 7, 2024, 3:42 PM

#

Let me check the knowledge base for you. One moment! :books:

#

I apologize for the confusion. If the SD15 model is failing with both 'fp32' and 'fp16' precision using the default VAE, it might be an issue specific to the SD15 model or its default VAE.

Here are a few more suggestions:

Try a Different VAE: If possible, try using a different VAE with the SD15 model. You can find different VAEs on the InvokeAI Huggingface repo (https://huggingface.co/InvokeAI).
Check for Updates: Make sure your InvokeAI and all models are up-to-date. Sometimes, these issues are resolved in newer versions.
Report the Issue: If none of the above solutions work, it would be best to report this issue on the InvokeAI GitHub Issues page (https://github.com/invoke-ai/InvokeAI/issues). Include all the details about your system, the version of InvokeAI you're using, and the exact steps you're taking when you encounter the issue. This will help the developers diagnose and fix the problem.

Remember to provide as much detail as possible to help the developers understand and resolve your issue.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@young mulch kapa.ai is still learning and improving, please let us know how it did by reacting below

young mulch Jan 7, 2024, 3:51 PM

#

Sorry, I made a mistake two responses ago. When I tried SDXL with sdxl-vae-fp16-fix, I had inadvertently deleted the IP Adapter. I tried it again with the IP Adapter, and got the black image again. So to summarize: I'm ALWAYS getting the black image with the IP Adapter, regardless of whether it's SD15 or SDXL, regardless of fp16 or fp32, or regardless of VAE. @viral radish

viral radishBOT Jan 7, 2024, 3:51 PM

#

Let me check the knowledge base for you. One moment! :books:

#

I see, so the issue of black images is consistently occurring when using the IP Adapter, regardless of the model, precision, or VAE used. This seems to be a more specific issue related to the IP Adapter itself.

Here are a few more suggestions:

Reinstall IP Adapter Models: As mentioned in the GitHub issue #5021 (https://github.com/invoke-ai/InvokeAI/issues/5021), deleting and redownloading the IP Adapter models resolved the issue for some users. Make sure you're downloading the correct models for your base model (SD15 or SDXL). You can find the models on the InvokeAI Huggingface repo (https://huggingface.co/InvokeAI).
Check for Errors: Look for any error messages or warnings in the console output when you run the model. This could provide clues as to what's going wrong.
Report the Issue: If none of these suggestions work, I recommend reporting the issue on the InvokeAI GitHub Issues page (https://github.com/invoke-ai/InvokeAI/issues). Include as much detail as possible, including your OS, GPU, VRAM, the version of InvokeAI you're using, and any error messages you're seeing. This will help the developers diagnose and fix the issue.

Remember to provide as much detail as possible to help the developers understand and resolve your issue.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@young mulch kapa.ai is still learning and improving, please let us know how it did by reacting below

young mulch Jan 7, 2024, 3:54 PM

#

I don't see any red text in the console indicating errors. I've already tried redownloading the IP Adapter, and I've tried with a couple of other IP Adapters as well. It just doesn't work with an IP Adapter. @viral radish

viral radishBOT Jan 7, 2024, 3:54 PM

#

Let me check the knowledge base for you. One moment! :books:

#

I understand your frustration. If you've already tried reinstalling the IP Adapter models and you're not seeing any error messages in the console, it's difficult to diagnose the issue without more information.

At this point, the best course of action would be to report this issue directly to the developers. You can do this by creating a new issue on the InvokeAI GitHub Issues page (https://github.com/invoke-ai/InvokeAI/issues).

When you create the issue, provide as much detail as possible:

Your operating system
GPU type
VRAM size
The version of InvokeAI you're using
The specific IP Adapter models you're using
The steps you're taking when you encounter the issue
Any other relevant information

This will help the developers understand the problem and work on a solution.

I'm sorry I couldn't provide a direct solution to your problem. AI and machine learning models can be complex and sometimes behave unexpectedly. The developers of InvokeAI are best equipped to help you resolve this issue.

#Black image issue when using IP Adapter

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

Alternatively, you could try using 'fp32' precision with the SD15 model and the default VAE, as this has been reported to work in some cases, as seen in GitHub issue #4083 (https://github.com/invoke-ai/InvokeAI/issues/4083).

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

Remember to provide as much detail as possible to help the developers understand and resolve your issue.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

Remember to provide as much detail as possible to help the developers understand and resolve your issue.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

I'm sorry I couldn't provide a direct solution to your problem. AI and machine learning models can be complex and sometimes behave unexpectedly. The developers of InvokeAI are best equipped to help you resolve this issue.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon: