This is a conversion of the Shuttle 3.1 Aesthetic model to NF4 Unet only 6.5GB (needs the T5_encoder, clip, an VAE)
https://huggingface.co/shuttleai/shuttle-3.1-aesthetic
This model get good quality images very quickly with just 4 steps (ideal for fixing details on the InvokeAI canvas)
It takes 18 seconds per image with RTX 3060 12GB
Thanks to the original author https://civarchive.com/user/shuttleai
Description
FAQ
Comments (2)
no me ale con esa calidad en forge alguna sugerencia?
That's strange, it works fine for me in Forge. Try setting Distilled CFG Scale to 1 - 2. Sampler: Euler. Schedule type: Simple. The result improves if you have 32gb of ram and use the t5xxl_fp16 encoder instead of the t5xxl_fp8
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.