Model Card: CherryFish_v5
🍒CherryFish_v5🐟
Alright, this is it. CherryFish_v5.
Another hopeless attempt to quiet the voices in my head before they finally drive me insane... i still don't think they're pleased.
My Method
My process remains a bit unconventional. It's born from a kind of sleep-deprived psychosis, chaotically using a mix of block weight merges and weighted differences until the aesthetic just feels right. It’s less of a science and more about chasing a specific feeling.
For those who like to dig into the details, I do my best to keep the metadata intact, so the recipe should be in there somewhere. 💕
🛠️ What's Different in v5?
This is the biggest leap forward yet.
The Ultimate Blend: This model perfectly marries the sharp, clean line work of anime with the sophisticated lighting of semi-realism. You get the best of both styles without having to fight the prompt.
Next-Level Lighting & Atmosphere: The way this version handles light is the most significant improvement. Volumetric light, neon glows, dramatic backlighting, and wet reflections are all baked in deep, giving scenes an incredible sense of mood.
Insane Material Rendering: The definition on materials is a huge step up. From the slick shine of latex and armor to the soft texture of a hoodie, everything just looks more tangible.
Less Bratty, More Intuitive: Remember how v4 had a bit of a rebellious streak? (e.g., writing "PLAIN WHITE" on a shirt you wanted plain). This version is much more cooperative. It listens to prompt weights more predictably, giving you better control without the sass.
🔧 Recommended settings:
The classic settings are still the way to go. No need to overcomplicate it.
Sampler: Euler a
Steps: 30-40 (30 is usually plenty).
CFG Scale: 4 - 6
Clip skip: 2
Hires. fix: Used on the showcase images but not completely necessary.
ADetailer is used on faces
Prompting Keywords & Tips
For the Core Style:
masterpiece, best quality, ultra-detailed, cinematic lighting, sharp focus, vibrant colors, atmospheric lighting, dynamic angle
For Character Focus:
1girl, solo, beautiful detailed face, beautiful detailed eyes, (white hair, pink tips), (black hair), choker, looking at viewer
Thank you to all the amazing creators whose models were sacrificed in my chaotic merging rituals to bring this to life. I couldn't have done it without your work. Enjoy
UPDATED | 🍒CherryFishv4🐟
🍒 CherryFishv3
Stylized. Saturated. Sharpened.
This version pushes way further into the anime side — brighter colors, sharper contrast, clean line work, and that crisp 2D energy. I steered away from semi-realism this time and leaned fully into stylized characters with a lot of punch. It’s built for vivid, expressive images that still play nice with detailed prompts.
Same as my other merges, i honestly don't really know everything thats in them.
too many all nighters have scramble my brain.
My method is a kind of sleep deprived psychosis chaotically utilizing a mix of block weights merges where i try to target the specific blocks controlling what i want to change as well as regular weighted difference checkpoint merging until they look how I want.
I do try to keep the metadata intact tho so if you are curious about the recipe it should be all there. 💕
🛠️ What's different from CherryFishv2:
Back to 2d anime (closer to cherry1)
Saturation and contrast are dialed in - everything just pops more, especially in neon-heavy or cinematic lighting.
Detail rendering feels tighter - especially on eyes, hair, and accessories.
Prompt responsiveness - is very good, tho it can give a little attitude from time to time. (i.e. trying to get a long sleeve t-shirt to not have any designs or logos on it so i crank up my "plain_white" tag to (plain_white:1:5) and the model says "FU" and writes in perfectly legible english "PLAIN WHITE" across the front of the shirt.) so yeah, a little bratty 🤷♀️
🔧 Recommended settings:
Sampler: Euler a
Steps: 30
CFG: 5.5–7 (tweak to taste)
Clip skip: 1 or 2 — doesn’t matter much in my experience
⬇️⬇️⬇️VERSION 2 CARD ⬇️⬇️⬇️
Just look at that big ass fish 😀
semi real or anime styles, just change where you put tags (3D and realistic, positive prompt for semi, negative prompt for anime)
Settings are same as always:
sample/scheduler - Euler a
steps - 30
guidance- 5.5/7ish
clip skip 1 or 2, i never really see much of a difference.
Models included in this mix are below in the suggested resources. Thank you to all the creators. 💕