Ludica is here!
I'm grateful for the GPU time from Redmond.AI that allowed me to finish this model!
This is a generalist model fine-tuned on Playground V2.
The model has a high capacity to generate realistic, artistic images, cars, people, and a wide variety of themes. It's a versatile model. Its a little improvement from already good PlaygroundV2 Model.
This model has been trained for approximately 90k steps.
I really hope you like the model and use it.
If you like the model and think it's worth it, you can make a donation to my Patreon or Ko-fi.
Follow me in my twitter to know before all about new models:
Description
FAQ
Comments (22)
what's the base model since it says "other?"
Title. "Playground V2".
@artificialguybr nvm, SDXL 1.0
@Strattenblitz It's actually no SDXL - "Playground" has the same architecture as SDXL but they trained it from the ground up so it's its own entity. HF has the base (not sure if it's on civitai). https://huggingface.co/playgroundai/playground-v2-1024px-aesthetic
@artificialguybr Very cool! Saw your tweet so came here. What did you use to train it software wise? Accelerate or something based on that?
@Strattenblitz Playground have same arch as SDXL 1.0. But its not the same model. But its traiend from the ground. Some XL things dont work.
@DSlater Kohya
Is this model not compatible with A1111 or does it need a specific VAE?
All I get is noisy images.
same
Does your A1111 work with other SDXL models? Ludica is fine-tuned on Playground V2, which uses the same architecture as SDXL.
@NoMansLand Yes, I use different SDXL based models without a problem.
How did you get it to train? I keep getting errors in kohya_ss.
This is the first model I've seen based on Playground v2. How many images did you use to train it?
You've posted a bf16 safetensor here -- was that intentional? Unexpected given that Playground v2 released in fp32 and fp16.
Hypothetically I should be able to run a model in bf16, but in practice I've never tested one in InvokeAI and I bet we have fp16 hardcoded someplace that will confuse it mightily when it tries to.
oh, that's right, you are one of the people who posts their models on HF in diffusers format! Thank you, that makes things much easier. =)
(and fwiw, you've posted fp32 and fp16 in diffusers format there, not bf16.)
Is it possible to add turbo and lcm
can you tell how to train
Can you please enable the model to allow it to be used for civitai's online image generation? Thank you for your consideration.
nice model, but how do I remove the image noise, the base playground v2 is not causing these issues, I'm using fooocus
Temporarily fixed the issue by putting the base playground v2 model as a very low strength refiner (Switch at 0.95)
does SDXL lora and controlnets works with this ?
I found something on Hugging Face that says no. Sometimes my attempts seem to change things, but I think it's more due to extra words in the prompt. Originally I thought using XL LoRA would generate nothing but noise. Maybe that was on a previous version of Auto1111. I now find I removed my few Playground models from my XY list for no reason. They're interesting models. Worth a look I think.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.