Type | |
Stats | 246 |
Reviews | (88) |
Published | May 15, 2023 |
Base Model | |
Training | Steps: 10,000 |
Hash | AutoV2 D59F39B747 |
This model is intented to be versatile enough to generate a wide variety of casual, NSFW, and landscape photos using a single checkpoint model!
Comparison with Popular Models
Below are the results from the x/y plot for comparison between other popular models like chillmix, grapefruit, and realisticvision:
This model appears to produce more natural looking images with a softer tone, this might be due to VAE usage. You can use the 840000 pruned VAE found here: https://huggingface.co/stabilityai/sd-vae-ft-mse
Recommended Parameters
Anime Sampler: DDIM + 2x hires 0.55 noise
Realistic Sampler: DPM++2M Karras
Sampling Steps: 15-25
Clip: 2
Negative Embeddings: versa-simple-neg
I don't have access to a desktop GPU so every sample image was generated on my laptop with an RTX2070 at resolutions no greater than 512x768 (before upscaling).