Sign In

SwedishFish

112
1.4k
50
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
473
0
1
Reviews
Published
Jan 25, 2025
Base Model
Illustrious
Usage Tips
Clip Skip: 2
Hash
AutoV2
AA17C0163C
default creator card background decoration
nolybutt's Avatar
nolybutt

SwedishFish v4 (Illustrious SDXL)

i cant tell if this is an improvement or not 🤷‍♀️ lol

Leave a comment and let me know! 💕


SwedishFish v3 (Illustrious SDXL)
Swedish Fish are still so yummy! 🐟

Overview:
SwedishFish are still as yummy as ever! 🐟 Version 3 builds upon the foundation of v2, refining and enhancing the balance between semi-realism and anime aesthetics. This version aims for even cleaner, more expressive outputs while improving adaptability across different styles. As always, it’s built from my favorite Illustrious checkpoints and LoRA merges. Thanks again to all the amazing creators. 💗

What's New in v3:
This update further fine-tunes SwedishFish’s style, enhancing consistency in facial features, lighting, and depth while keeping its signature flexibility. Expect sharper details, smoother rendering, and improved coherence across different prompt styles. It also handles natural language prompting even better while maintaining its love for tag-heavy approaches.


SwedishFish v2 (Illustrious SDXL)
Swedish Fish are still so yummy! 🐟

Overview:
SwedishFish are still as yummy as ever! 🐟 Version 2 continues to build on the versatility and style of the first version, pushing the model’s ability to handle semi-realistic and anime-flared artwork even further. This is achieved through the merging of my favorite Illustrious checkpoints and LoRA. Thanks again to all the amazing creators. 💗

What's New in v2:
This version focuses on refining the overall style and adaptability of SwedishFish v1, improving detail, expressiveness, and flexibility across different artistic styles and themes.

What It’s Good At:

  • NSFW & SFW: Performs well in both without LoRAs, but it also integrates smoothly with LoRA add-ons.

  • Prompting: Improved natural language prompt adherence over version 1, but still prefers tag-heavy prompts. (Sometimes I use a combination of both depending on what I'm trying to have the model do. Just mess around with it and see what works.) Tags like "masterpiece" and "best quality" are helpful but not necessary—use whichever quality tags you like, as long as you include some kind of quality tag.

  • Styles: Achieve semi-realistic or anime aesthetics based on prompt adjustments (e.g., add "3D/realistic" to positives for semi-real; add them to negatives for anime).

Suggested Settings:

  • Sampler: Euler a

  • Steps: 30

  • CFG (Guidance): 5.5 to 7

  • Clip Skip: 1 or 2 (user preference)

  • Resolution: 768×1344 (Width × Height)

The images in the showcase all use ESRGAN-UltraSharp-4x for HI-RES FIX, but this is absolutely not necessary. Images still look great without Hi-Res.

I recommend using adetailer for faces.


IF YOU LIKE THE MODEL PLEASE GIVE IT A THUMBS UP. AND AS ALWAYS, POST YOUR IMAGES! HAVE FUN! K BYE!💕


Swedish Fish v1 Background:
This merge is Cherry Fish, but mixed with another merge I made a week or two ago (no clue what I used for it 🤷‍♀️). I combined the random merge with Cherry Fish using the Smooth Step Block Weight Merge preset.

It’s pretty good at whatever you wanna do with it! Great NSFW without any LoRA use, but also works well with LoRAs if you choose.

Prompting Notes:

  • Likes tag-heavy prompts. Natural language works, but tagging performs better.

  • Semi-real or anime styles—just change where you place the tags (e.g., use "3D and realistic" in positives for semi-real and in negatives for anime).

Settings (Same as Always):

  • Sampler/Scheduler: Euler a

  • Steps: 30

  • Guidance: 5.5 to 7ish

  • Clip Skip: 1 or 2 (I never really see much difference)

Models included in this mix are listed below under suggested resources.
Thank you to all the creators! 💕