Popular Stable Diffusion Models Overview (Part 2)

This article provides a comprehensive overview of commonly used Stable Diffusion Webui models, including detailed information about fantasy-world, pony diffusion, Openjourney v4, dosmix, meinamix, seek.art MEGA, Pastel Boys, and AnyLoRA models.

Fantasy-World Model by Satyam_SSJ10

Satyam_SSJ10/fantasy-world is a fine-tuned version of the original Dpepteahand3 model. Here are the key features:

1. Fine-tuned using artwork from various concept artists
2. The artist spent over two weeks collecting and manually curating artworks
3. Generated images will display the Dpepteahand3 model name, which can be ignored

Pony Diffusion V4 Model by AstraliteHeart

AstraliteHeart/pony-diffusion-v4 is a specialized model for generating pony and plush-style images. Key features:

• Fine-tuned using 3 million text-image pairs from derpibooru, e621, and danbooru
• Trained for 15 epochs with a 5.0e-6 learning rate
• Recommended to use secondary CLIP and add derpibooru_p_95 to prompts and derpibooru_p_low to negative prompts

Openjourney V4 Model

prompthero/openjourney-v4 is a versatile model with the following characteristics:

• Trained on 124,000 images for 12,400 steps and 4 epochs (32 hours total)
• No longer requires the mdjrny-v4 style prompt
• Supports multiple spatial styles (conference rooms, offices, warehouses, etc.)
• Based on Stable Diffusion 1.5 Find more prompts at Prompthero

Dosmix Model

DiaryOfSta/dosmix offers these optimization tips:

• Use (Realistic:0.1~1.4) or (realistic:0.1~1) in negative prompts
• Recommended VAE: kl-f8-anime2 or vae-ft-mse-840000-ema-pruned
• Use 0.5 denoising and 2x upscaling
• Best for upper body and bust shots

Meinamix Model

Meina/meinamix is a merged model combining:

• MeinaMix V1-6
• MeinaPastel V3
• MeinaHentai V2
• Night Sky YOZORA Style Model
• PastelMix
• Facebomb

Recommended settings:

• Samplers: Euler a (40-60 steps) or DPM++ SDE Karras (30-60 steps)
• CFG Scale: 7
• Portrait resolutions: 512x768, 512x1024
• Landscape resolutions: 768x512, 1024x512, 1536x512
• Hires.fix: R-ESRGAN 4x+Anime6b
• Clip Skip: 2

Seek.art MEGA Model

coreco/seekart-mega is a versatile “all-in-one” model that outperforms SD 1.5 in multiple styles. Best practices:

• Use resolutions above 640px
• Apply vae-ft-mse-840000-ema VAE
• Include negative prompts to exclude unwanted styles
• Be specific in prompts
• Works well with sd-dynamic-thresholding

Pastel Boys Model

bodlo/pastelboys2d specializes in male character generation with:

• Brighter colors and detailed backgrounds
• Recommended settings:
  • Sampling: Eular a / DPM++ SDE Karras
  • Clip skip: 2
  • Hires.fix upscaler: R-ESRGAN 4x+Anime6B
  • CFG Scale: 7-9
  • VAE: vae-ft-mse-840000-ema-pruned / kl-f8-anime2

AnyLoRA Model

Lykon/anylora is a checkpoint model designed for:

• Enhanced LoRA training compatibility
• Neutral style base for accurate LoRA application
• Recommended weight of 0.65, adjustable to 1
• Efficient memory usage
• Ideal for both training and inference
• Perfect for Colab and local environments