Gravy Mix is still Great! but it has been superseded by PhotoSomnia.
This blend started as a mess of mashing models together but has become a very delicately balanced and versatile model. The objective has been to improve photorealism and hand/face details.
I've found it to be great for sfw and erotic female fake photos, idk how it is for porn or other subjects that's not my niche. It can also get quite artistic with creative prompting.
While Gravy Mix 1 is still somewhere at the heart of things, It's undeniable that Dreamlike Photoreal in particular has made a huge difference and helped to make this quality of output possible more than any other contributing model. Their modified licensing requirements as visible on their page need to be read, understood and complied with and will affect commercial applications, but this model is very suitable for creative use by hobbyists and independent professionals. Otherwise permissions for use aren't mine to give as this is just a blend of other people's work, but I'd appreciate being given credit if these models are being used in blends.
Description
Was trying to re-create an intermediate blend and instead I came up with something even better, it's worthy of a new, final version. It's an improvement on 2.4's composition but with better photorealism for human subjects than that or 2.3, at the expense of versatility.
The more complex the prompt, the higher cfg up to 10 is recommended, with 4 as a minimum for the simplest of prompts. The larger the pic the better the details but composition starts to suffer above 768px height, lower resolution images work well as an input for img2img at higher resolutions, or the automatic1111 hires fix will do it for you.
This was made by building on 2.4 and remixing with some previously used models, plus TheAlly's Mix II: Churned.
It has vae-ft-mse-840000 baked in. Don't forget the config file!
I'm calling it a day after this one, it's been fun trying things out and finding some things that work and a lot that don't, but the returns on improved quality for the time spent are diminishing so I'll be focusing on other projects. Please enjoy this gift to the AI art community and use it to bring to life a little more beauty in this world.
FAQ
Comments (3)
Hi, im using Invoke AI and i have problems with config file when i try to load the model.
Error:
model gravyMixPhotorealism_v3final could not be loaded: LatentDiffusion.__init__() missing 1 required positional argument: 'personalization_config'
Traceback (most recent call last):
File "C:\Users\myname\invokeai\.venv\lib\site-packages\ldm\generate.py", line 956, in set_model
model_data = cache.get_model(model_name)
File "C:\Users\myname\invokeai\.venv\lib\site-packages\ldm\invoke\model_manager.py", line 119, in get_model
requested_model, width, height, hash = self._load_model(model_name)
File "C:\Users\myname\invokeai\.venv\lib\site-packages\ldm\invoke\model_manager.py", line 349, in loadmodel
model, width, height, model_hash = self._load_ckpt_model(
File "C:\Users\myname\invokeai\.venv\lib\site-packages\ldm\invoke\model_manager.py", line 450, in loadckpt_model
model = instantiate_from_config(omega_config.model)
File "C:\Users\myname\invokeai\.venv\lib\site-packages\ldm\util.py", line 93, in instantiate_from_config
return get_obj_from_str(config["target"])(**config.get("params", dict()), **kwargs)
TypeError: LatentDiffusion.__init__() missing 1 required positional argument: 'personalization_config'
I haven't tried out InvokeAI in a long time, Invoke is great for artistic workflow but I'm using Automatic1111 because of it's easy model blending function and it's easy to test out blends quickly.
A couple of things I can think of without really knowing what the problem is:
1. try loading the model without the config, it's not that much worse without it as it's only a hangover from blending in one of the many models used.
2. check all the filenames/paths and make sure they match the actual files.
I have seen other people getting a similar error when trying to load SD2.1 so the config may be just as incompatible despite being for an SD1.5 model. I'm really not that familiar with InvokeAI and utterly lack in-depth understanding of its model loading process. Maybe someone with basic knowledge of what the code does can explain what the problem might actually be but I don't have time to troubleshoot this, sorry.
Good luck!
@grvy Tried your suggestions but no luck. 1 doesn´t work, it gives me back a brown image. 2 was thouroughly checked yesterday.
Since i was already thinking in setting up an Automatic1111 install to use some features like LORA that invoke doesn´t have yet, i installed it and it works! Where do i have to put the config file in Automatic? Same folder of the model?. Thanks for answering back!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.