Add Raykreate Krea 2 Turbo models and refresh model card

Files changed (12) hide show

.gitattributes +8 -0
KREA2_TURBO/Raykreate_v1.0.safetensors +3 -0
KREA2_TURBO/Raykreate_v3.1_INT8_CONVROT.safetensors +3 -0
README.md +394 -357
_IMG_/ALL_00007_.png +3 -0
_IMG_/ALL_00007_preview.png +3 -0
_IMG_/HF_KREA2-TURBO.png +0 -0
_IMG_/HF_KREA2-TURBO.svg +1 -0
_IMG_/Raykreatev1_00001_.png +3 -0
_IMG_/Raykreatev1_00002_.png +3 -0
_IMG_/Raykreatev3_00001_.png +3 -0
_IMG_/Raykreatev3_00002_.png +3 -0

.gitattributes CHANGED Viewed

@@ -64,3 +64,11 @@ _WORKFLOWS/Raymnants-Rayctifier-Rayburn_Workflow.png filter=lfs diff=lfs merge=l
 _WORKFLOWS/Rayqwest_Workflow.png filter=lfs diff=lfs merge=lfs -text
 _WORKFLOWS/Rayzist_Workflow.png filter=lfs diff=lfs merge=lfs -text
 _IMG_/ALL_00006_.png filter=lfs diff=lfs merge=lfs -text

 _WORKFLOWS/Rayqwest_Workflow.png filter=lfs diff=lfs merge=lfs -text
 _WORKFLOWS/Rayzist_Workflow.png filter=lfs diff=lfs merge=lfs -text
 _IMG_/ALL_00006_.png filter=lfs diff=lfs merge=lfs -text
+KREA2_TURBO/Raykreate_v1.0.safetensors filter=lfs diff=lfs merge=lfs -text
+KREA2_TURBO/Raykreate_v3.1_INT8_CONVROT.safetensors filter=lfs diff=lfs merge=lfs -text
+_IMG_/ALL_00007_.png filter=lfs diff=lfs merge=lfs -text
+_IMG_/ALL_00007_preview.png filter=lfs diff=lfs merge=lfs -text
+_IMG_/Raykreatev1_00001_.png filter=lfs diff=lfs merge=lfs -text
+_IMG_/Raykreatev1_00002_.png filter=lfs diff=lfs merge=lfs -text
+_IMG_/Raykreatev3_00001_.png filter=lfs diff=lfs merge=lfs -text
+_IMG_/Raykreatev3_00002_.png filter=lfs diff=lfs merge=lfs -text

KREA2_TURBO/Raykreate_v1.0.safetensors ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:902e1b2cf44c9c6206cc60e6257e3082291c284788d99caa0b6f08d58f035325
+size 12820143220

KREA2_TURBO/Raykreate_v3.1_INT8_CONVROT.safetensors ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:a6087f6bf2b7e9e6ec6983932dd13e0eb942c30b7fc1c16d90a34cef157628a0
+size 13492097280

README.md CHANGED Viewed

@@ -1,357 +1,394 @@
----
-license: apache-2.0
-language:
-  - en
-base_model:
-  - Tongyi-MAI/Z-Image-Turbo
-  - black-forest-labs/FLUX.1-dev
-  - lodestones/Chroma1-Base
-  - stabilityai/stable-diffusion-xl-base-1.0
-  - Qwen/Qwen-Image
-  - black-forest-labs/FLUX.1-Krea-dev
-tags:
-  - model-zoo
-  - generative-models
-  - flux
-  - sdxl
-  - chroma
-  - qwen-image
-  - krea
-  - z-image
-  - photographic
-  - analog
-  - creative-ai
-  - ray
-  - ray-models
-library_name: diffusers
-pipeline_tag: text-to-image
----
-# Ray’s Models Zoo
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/RAY_LOGOT.png" width="100">
-An AI repository of curated visual styles, a personal generative image models collection.</br>
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv1_00001_.png" target="_blank">
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/ALL_00006_.png" width="500">
-</a>
----
-## TL;DR
-A curated collection of custom-trained generative models built on personal assets across Flux-Dev, Krea, SDXL, Chroma, Qwen-Image, Z-Image, and others — all designed to deliver my own distinctive visual style, mostly photographic, but not exclusively.
-<br>
-I also created a small standalone inference app that wraps my Z-image_Turbo model Rayzist here:
-<a href="https://github.com/MutantSparrow/JustRayzist" target="_blank">https://github.com/MutantSparrow/JustRayzist</a>
-## Model Philosophy
-Across the different families (SDXL, Flux, Qwen-Image, Chroma, Z-Image), the intention remains the same.
-They were trained on a **unique, hand curated dataset** built around years of original street photography and later enhanced with synthetic data from my own models.
-My overarching goal is to capture a **natural, analog-leaning photographic sensibility** without sacrificing creative range. Many of the models inherit a subtle film-grain texture, dimmed contrast, and an overall organic feel. Others explore alternate aesthetics — oil painting, digital illustration, stylized comic rendering — but always through the lens of my own visual taste.
-These models are designed for **everyone**.
-Hopefully you'll like some of them.
----
-## Model Families Overview
-Some families are specialized — like the SDXL branch, with portrait-centric Raymnants, painterly Rayctifier, and comic-styled Rayburn — while newer architectures such as Flux, Qwen-Image, and Z-Image tend to be more powerful generalists with a strong stylistic backbone.
-It's also a proof-of-concept for **blockwise LoRA merging** as a method for finetuning models rather than full training. Besides my SDXL models, most of the checkpoints here were done with the following process:
--  Train a LoRA on a specific concept (style or subject)
--  Turn the LoRA influence to 0 and turn it back on on every block of the models one by one to generate a control image
--  Identify the layers/blocks most affected by the concepts we want to reinforce
--  tune the values only in thosr blocks by checking the control images until satisfactory
--  bake weights in the model
--  wash, rinse, repeat with a new LoRA
-A typical vanilla-to-v1 transition takes usually 4-8 LoRAs. I oftentimes merge LoRAs thematically on a single model, but keep multiple model branches in parallel before merging them together (e.g. one Photo style branch, one character detail branch, etc.)
-## Model List
-- **CHROMA → Raydiant**  Lightweight analog-photography oriented experiments, built on Chroma v48. Instable but creative
-- **FLUX → Rayflux**  Flagship generalist model line with analog grain, strong realism, and expressive control.
-- **FLUX_KREA → Rayflux Krea**  Experimental Flux-Krea refinement with excellent skin rendering and strong style.
-- **QWEN_IMAGE → RayQwest**  Qwen-Image based models that remove plasticness and restore analog grit and Ray-style texture.
-- **SDXL → Raymnants + Rayctifier + Rayburn**  SDXL models spanning portrait realism, oil painting, and stylized comic style.
-- **Z-IMAGE → Rayzist**   Fast Turbo model that performs above typical distilled models expectations, with good text rendering to boot.
-SDXL models will be fp16, ZIT bf16, all the rest is quantized to fp8.
----
-## Recommended Models Lines
-1. **Rayflux** v1, Photoplus and v3
-2. **RayQwest**
-3. **Raymnants** 3.0 for best balance
-4. **Rayzist**
-Other models all have their strengths but if you just gotta make a choice, start with the above
----
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_FLUX.png">
-> **Baseline Rayflux settings (I recommend you experiment):**
-> Sampler: **Euler/Beta** · Guidance: **2.5** · Steps: **16** · min/max Shift: **1/1.5**
----
-## Rayflux v1.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv1_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv1_00002_.png"
-       alt="Rayflux preview"
-       style="max-width:100%; height:auto;">
-</a>
-Less photorealistic than Photoplus but more vibrant and creatively expressive. A versatile all-rounder that works well across photography, digital illustration, and hybrid “photo-illustration” styles.
-## Rayflux Photoplus
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv2_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv2_00002_.png"
-       alt="Photoplus preview"
-       style="max-width:100%; height:auto;">
-</a>
-AKA Rayflux v2. A best-in-class photorealistic Rayflux variant with excellent analog-style grain, stable anatomy, and a broad creative range. Ideal for portraits, street scenes, and detailed natural photography with a cinematic twist.
-## Rayflux v3.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv3_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv3_00002_.png"
-       alt="Rayflux v3.0 preview"
-       style="max-width:100%; height:auto;">
-</a>
-Trained on the same dataset as Photoplus enhanced with outputs from v1. Strikes a balanced hybrid identity: clean photographic detail with meaningful creative flexibility and a slightly more mature, cohesive visual language.
-## Rayflux Horndog
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/RayfluxvX_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/RayfluxvX_00002_.png"
-       alt="Horndog preview"
-       style="max-width:100%; height:auto;">
-</a>
-A model I overtuned on anatomy trying to improve on Photoplus; it tends to produce accidental NSFW content so don't forget to prompt for clothing! High detail, vivid rendering, and strong contrast, but is harder to control than previous models. Suffers a bit from same-face syndrome (use LoRAs to avoid that).
-## Rayflux v3.x → v4
-A consolidated series of refinement experiments designed to progressively “de-hornify” Horndog — cleaner anatomy, fewer unintended biases, and improved stability — while preserving the core Rayflux style. In fine, they offer different flavors than v1-v3 but failed to surpass them.
----
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_QWEN_IMAGE.png">
-> **Baseline Rayqwest settings (I recommend you experiment):**
-> Sampler: **Euler/Simple** · Steps: **16**
-## RayQwest v1.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayqwest_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayqwest_00002_.png"
-       alt="RayQwest preview"
-       style="max-width:100%; height:auto;">
-</a>
-RayQwest v1.0 is designed to remove the characteristic “AI plasticness” often seen in Qwen-Image outputs, restoring a more natural analog grit and the recognizable Ray-style aesthetic. It maintains excellent prompt comprehension, handles a wide range of styles with ease, and runs fast — all while giving Qwen’s already powerful architecture a stronger artistic backbone. Carries some of Qwen’s usual softness — less sharp than Rayflux — but compensates with **excellent prompt accuracy and expressive range**. Works well for portraits, lifestyle photography, stylized art, and creative compositions.
-## RayQwest v0.7
-RayQwest v0.7 is a strong pre-v1 iteration characterized by a heavier analog feel, noticeable grain, and a slightly flatter dynamic range. It’s a bit noisier than v1.0, but compensates with **excellent aesthetic vibes**, maintaining solid prompt comprehension and creative flexibility. Despite being an intermediate version, it remains highly usable and flavorful.
-## RayQwest v0.65
-RayQwest v0.65 was trained on a **different dataset** entirely, giving it a more **dreamlike**, **saturated**, and softer look. It carries a unique atmosphere — less sharp than v0.7 and v1.0, but with an appealing ethereal quality that makes it stand apart. A charming early version that still holds its own artistically.
----
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_SDXL.png">
-> **Baseline SDXL settings (I recommend you experiment):**
-> Sampler: **DPMPP2M/Karras** · cfg:**5-7** · Steps: **20-30**
-## Raymnants 3.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raymnants_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raymnants_00002_.png"
-       alt="Raymnants preview"
-       style="max-width:100%; height:auto;">
-</a>
-Realistic model with a good sense for Art. Great at portraits, wide ethnic and body type range. The model that started it all. What would happen if I trained a bunch of street photos into an image model? The answer was: RAYMNANTS The rudimentary methodology and complete lack of insight make it nothing short of a miracle that it came out that good. Still took half a dozen iterations before 3.0 though!
-## Rayctifier 1.5
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayctifier_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayctifier_00002_.png"
-       alt="Rayctifier preview"
-       style="max-width:100%; height:auto;">
-</a>
-My artistic model, geared towards painting style and all around illustrations, with a classical painting flair.
-I’ve finetuned a branch of my realistic model, RAYMNANTS, on about 600 public domain paintings from the Smithsonian online archive, concentrating on the 1750-1850 period.
-The result is a model that absorbed some of the style from those classical painting while retaining above average photo capacity. It will perform really well for any prompt destined to output a painting, do well on illustration, but generally underperforms in photo-style context
-## Rayburn 1.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayburn_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayburn_00002_.png"
-       alt="Rayburn preview"
-       style="max-width:100%; height:auto;">
-</a>
-My overly opinionated stylized model, geared comic-style illustrations, it will take everything you thrown at it, and give it back, with a twist. I’ve finetuned a branch of my painterly model, RAYCTIFIER, on a mix of about 800 hand-drawn illustrations and synthetic data.
-Careful cross-breeding to cultivate some style traits resulted in a very bold style you can’t really prompt out of but that has a lot of personality. It’s seriously overfit, but that’s what makes it the way it is. The model will perform well at taking its own stylistic stance to any kind of prompt you’ll thrown at it. It’s great fun, and also a bit wild!
-## Rayctifier 1.0
-Rayctifier 1.0 was an earlier version of the model that I thought wasn't deep enough into the painting style I wanted for it. But if you personally find Rayctifier 1.5 is a bit too much, then 1.0 might be just your thing.
-## Raymnants 3.0 LightningW 4L1.0 & 4L2.0
-Lightning distilled variants from Raymnants 3.0, they aimed at accelerating sampling while preserving as much of the original style as possible. YMMV!
-Both versions went through a different distillation process.
-- L1 is closer to the 3.0 look but thinner in detail and nuance.
-> Sampler: **Euler/sgm_uniform** · cfg:**1** · Steps: **8**
-- L2 is more aggressively distilled; punchier but also more chaotic and further from base 3.0.
-> Sampler: **DPMPPSDE/sgm_uniform** · cfg:**2** · Steps: **6**
-## Raymnants 3.4 → 6.2 — Iterative Evolution Line
-Raymnants 3.4 through 6.2 form an exploratory lineage built on 3.0. At the time, the training process involved a lot of experimentation, resulting in stylistic “yoyo effects” between versions — sometimes more realistic, sometimes more expressive, sometimes more saturated.
-At the end of this fluctuating path, the 6.x models stand out as flexible, bright, saturated, and artistically dynamic, if slightly less realistic than 3.0.
-## RayburnBlitz 1.0
-RayburnBlitz 1.0 is the distilled, lightning-fast variant of Rayburn. It remains stylized and comic-driven but loses some of the original model’s extreme boldness that was achieved through overfitting and lost through distillation.
-> Sampler: **Gradient_Estimation/beta57** · cfg:**1.5** · Steps: **6**
----
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_Z-IMAGE-TURBO.png">
-> **Baseline Rayzist settings (I recommend you experiment):**
-> Sampler: **Euler/Simple** · cfg: **1** · Steps: **8**
-## Rayzist v1.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayzist_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayzist_00002_.png"
-       alt="Rayzist preview"
-       style="max-width:100%; height:auto;">
-</a>
-Rayzist v1.0 is a fast, generalist Z-Image Turbo model with unusually strong prompt adhesion for its class. Despite its distilled nature, it's mostly uncensored, has a broad conceptual range, performs well across photography, anime, UI and comic-style rendering. It is also ridiculously good at generating readable text inside the image, which is unheard of for Turbo-tier models. Rapid ideation, multi-style image generation, but 8GB VRAM workflows, photography and anime rendering, and text-in-image generation: Rayzist has it all. Well, almost: it is still a distilled Turbo model and some visual concepts simply aren’t well encoded (e.g., cyclops / one-eyed characters). These gaps are unpredictable and usually only discovered through experimentation.
----
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_CHROMA.png">
-> **Baseline Raydiant settings (I recommend you experiment):**
-> Sampler: **Euler/Beta** · cfg:**4.5** · Steps: **16**
-> **Notes:** Pad the T5 tokenizer (min_length = 3); works well with dedicated negative prompting.
-## Raydiant v1.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raydiant_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raydiant_00002_.png"
-       alt="Raydiant preview"
-       style="max-width:100%; height:auto;">
-</a>
-Raydiant v1.0 is a Chroma-based model spun off from **Chroma v48**, tuned to produce uncensored analog-style photography with strong contrast, light fuzziness, and a natural film-grain feel. Despite its photographic backbone, it also performs surprisingly well across bold illustrated genres, maintaining expressive color and stylistic weight. It's quite hard to use, with islands of stability in the concepts where it performs well and other places where it's just falling flat on its face. I struggled a lot to train the model and never went back after Chroma v48, but I still like Raydiant for all its tantrums: bleeding concepts, garbled latents and unpolished hands/feet.
----
-<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_FLUX-KREA.png">
-> **Baseline Rayflux Krea settings (I recommend you experiment):**
-> Sampler: **ER_SDE/Beta** · Guidance: **3** · Steps: **20** · min/max Shift: **0.5/2**
-## Rayflux Krea v1.0
-<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayflux-Krea_00001_.png" target="_blank">
-  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayflux-Krea_00002_.png"
-       alt="Rayflux-Krea preview"
-       style="max-width:100%; height:auto;">
-</a>
-Rayflux Krea v1.0 is a photo style checkpoint tuned for **dreamy, stylish pictures** and a bold flavorful look. It carries a distinct artistic flair and handles a wide variety of styles well, from creative photography to expressive illustration. Compared to standard Rayflux models, it has its own flavor — visually more opinionated than vanilla Flux, but also slightly more resistant to absorbing new concepts. It does suffer a bit from same-face syndrome though, and have some holes in the concepts it covers.
----
-## Repository Structure
-```text
-CHROMA/
-    Raydiant v1.0
-FLUX/
-    Rayflux Horndog
-    Rayflux Photoplus
-    Rayflux v1.0
-    Rayflux v1.0 (UNet-only)
-    Rayflux v3.0
-    Rayflux v3.x (3.1–3.4)
-    Rayflux v4.0
-FLUX_KREA/
-    Rayflux Krea v1.0
-QWEN_IMAGE/
-    RayQwest v0.65
-    RayQwest v0.7
-    RayQwest v1.0
-SDXL/
-    Rayburn 1.0
-    RayburnBlitz 1.0
-    Rayctifier 1.0
-    Rayctifier 1.5
-    Raymnants 3.0
-    Raymnants 3.0 LightningW 4L1.0
-    Raymnants 3.0 LightningW 4L2.0
-    Raymnants 3.4
-    Raymnants 3.4b
-    Raymnants 4.0
-    Raymnants 4.1
-    Raymnants 5.0
-    Raymnants 6.0
-    Raymnants 6.1
-    Raymnants 6.2
-Z-IMAGE/
-    Rayzist v1.0
-_IMG_/
-   (all reference images go here)
-_WORKFLOWS/
-   (added ComfyUI workflows references)
-```
----
-## License
-All models in this zoo are trained on personal and/or mixed-source data and **inherit the license and usage constraints of their respective base models** (see metadata above). Check the upstream repositories for detailed license terms.
----
-## Credits
-All models, datasets, training, and curation by **Ray / MutantSparrow**.

+---
+license: apache-2.0
+language:
+  - en
+base_model:
+  - krea/Krea-2-Turbo
+  - Tongyi-MAI/Z-Image-Turbo
+  - black-forest-labs/FLUX.1-dev
+  - lodestones/Chroma1-Base
+  - stabilityai/stable-diffusion-xl-base-1.0
+  - Qwen/Qwen-Image
+  - black-forest-labs/FLUX.1-Krea-dev
+tags:
+  - model-zoo
+  - generative-models
+  - flux
+  - sdxl
+  - chroma
+  - qwen-image
+  - krea
+  - krea2-turbo
+  - z-image
+  - photographic
+  - analog
+  - creative-ai
+  - ray
+  - ray-models
+library_name: diffusers
+pipeline_tag: text-to-image
+---
+# Ray’s Models Zoo
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/RAY_LOGOT.png" width="100">
+An AI repository of curated visual styles, a personal generative image models collection.</br>
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/ALL_00007_.png" target="_blank">
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/ALL_00007_preview.png" width="500" alt="Ray model zoo cover">
+</a>
+---
+## TL;DR
+A curated collection of custom-trained generative models built on personal assets across Flux-Dev, Krea, SDXL, Chroma, Qwen-Image, Z-Image, and others — all designed to deliver my own distinctive visual style, mostly photographic, but not exclusively.
+<br>
+I also created a small standalone inference app that wraps my Z-image_Turbo model Rayzist here:
+<a href="https://github.com/MutantSparrow/JustRayzist" target="_blank">https://github.com/MutantSparrow/JustRayzist</a>
+## Model Philosophy
+Across the different families (SDXL, Flux, Qwen-Image, Chroma, Z-Image), the intention remains the same.
+They were trained on a **unique, hand curated dataset** built around years of original street photography and later enhanced with synthetic data from my own models.
+My overarching goal is to capture a **natural, analog-leaning photographic sensibility** without sacrificing creative range. Many of the models inherit a subtle film-grain texture, dimmed contrast, and an overall organic feel. Others explore alternate aesthetics — oil painting, digital illustration, stylized comic rendering — but always through the lens of my own visual taste.
+These models are designed for **everyone**.
+Hopefully you'll like some of them.
+---
+## Model Families Overview
+Some families are specialized — like the SDXL branch, with portrait-centric Raymnants, painterly Rayctifier, and comic-styled Rayburn — while newer architectures such as Flux, Qwen-Image, and Z-Image tend to be more powerful generalists with a strong stylistic backbone.
+It's also a proof-of-concept for **blockwise LoRA merging** as a method for finetuning models rather than full training. Besides my SDXL models, most of the checkpoints here were done with the following process:
+-  Train a LoRA on a specific concept (style or subject)
+-  Turn the LoRA influence to 0 and turn it back on on every block of the models one by one to generate a control image
+-  Identify the layers/blocks most affected by the concepts we want to reinforce
+-  tune the values only in thosr blocks by checking the control images until satisfactory
+-  bake weights in the model
+-  wash, rinse, repeat with a new LoRA
+A typical vanilla-to-v1 transition takes usually 4-8 LoRAs. I oftentimes merge LoRAs thematically on a single model, but keep multiple model branches in parallel before merging them together (e.g. one Photo style branch, one character detail branch, etc.)
+## Model List
+- **KREA2_TURBO → Raykreate**  [RayKreate v3.1 INT8 CONVROT](https://huggingface.co/MutantSparrow/Ray/resolve/main/KREA2_TURBO/Raykreate_v3.1_INT8_CONVROT.safetensors) and [Raykreate v1.0](https://huggingface.co/MutantSparrow/Ray/resolve/main/KREA2_TURBO/Raykreate_v1.0.safetensors). Baseline: Euler/Simple, 8 steps, CFG 1.
+- **CHROMA → Raydiant**  Lightweight analog-photography oriented experiments, built on Chroma v48. Instable but creative
+- **FLUX → Rayflux**  Flagship generalist model line with analog grain, strong realism, and expressive control.
+- **FLUX_KREA → Rayflux Krea**  Experimental Flux-Krea refinement with excellent skin rendering and strong style.
+- **QWEN_IMAGE → RayQwest**  Qwen-Image based models that remove plasticness and restore analog grit and Ray-style texture.
+- **SDXL → Raymnants + Rayctifier + Rayburn**  SDXL models spanning portrait realism, oil painting, and stylized comic style.
+- **Z-IMAGE → Rayzist**   Fast Turbo model that performs above typical distilled models expectations, with good text rendering to boot.
+SDXL models will be fp16, ZIT bf16, RayKreate v3.1 is INT8 CONVROT, and all the rest is quantized to fp8.
+---
+## Recommended Models Lines
+1. **Raykreate** v3.1, v1.0
+2. **Rayzist**
+3. **Rayflux** v1, Photoplus and v3
+4. **Rayqwest**
+Other models all have their strengths but if you just gotta make a choice, start with the above
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_KREA2-TURBO.png" alt="xKREA2 TURBO">
+> **Baseline Krea 2 Turbo settings (I recommend you experiment):**
+> Sampler: **Euler/Simple** · Steps: **8** · CFG: **1**
+---
+## RayKreate v3.1 INT8 CONVROT
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raykreatev3_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raykreatev3_00002_.png"
+       alt="RayKreate v3.1 preview"
+       style="max-width:100%; height:auto;">
+</a>
+[Download RayKreate v3.1 INT8 CONVROT](https://huggingface.co/MutantSparrow/Ray/resolve/main/KREA2_TURBO/Raykreate_v3.1_INT8_CONVROT.safetensors)
+The best of both worlds: more creativity, good photographic style, and more sensible behavior. It's really a great model. As an INT8 variant, it's also super lightweight and fast on modern GPUs. It's pretty good at anime, concept art, fantasy illustrations, pseudo-3D renders, and more. Give it a try!
+## Raykreate v1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raykreatev1_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raykreatev1_00002_.png"
+       alt="Raykreate v1.0 preview"
+       style="max-width:100%; height:auto;">
+</a>
+[Download Raykreate v1.0](https://huggingface.co/MutantSparrow/Ray/resolve/main/KREA2_TURBO/Raykreate_v1.0.safetensors)
+My best photorealistic model so far. It sometimes lacks creativity, but its prompt understanding is leaps and bounds above previous-generation models. It's great at just about everything photographic. In the process of making it more flexible, I've abliterated the built-in NSFW refusal layer, so the model can sometimes be a bit too eager. It also isn't the best at illustrative styles.
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_FLUX.png">
+> **Baseline Rayflux settings (I recommend you experiment):**
+> Sampler: **Euler/Beta** · Guidance: **2.5** · Steps: **16** · min/max Shift: **1/1.5**
+---
+## Rayflux v1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv1_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv1_00002_.png"
+       alt="Rayflux preview"
+       style="max-width:100%; height:auto;">
+</a>
+Less photorealistic than Photoplus but more vibrant and creatively expressive. A versatile all-rounder that works well across photography, digital illustration, and hybrid “photo-illustration” styles.
+## Rayflux Photoplus
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv2_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv2_00002_.png"
+       alt="Photoplus preview"
+       style="max-width:100%; height:auto;">
+</a>
+AKA Rayflux v2. A best-in-class photorealistic Rayflux variant with excellent analog-style grain, stable anatomy, and a broad creative range. Ideal for portraits, street scenes, and detailed natural photography with a cinematic twist.
+## Rayflux v3.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv3_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayfluxv3_00002_.png"
+       alt="Rayflux v3.0 preview"
+       style="max-width:100%; height:auto;">
+</a>
+Trained on the same dataset as Photoplus enhanced with outputs from v1. Strikes a balanced hybrid identity: clean photographic detail with meaningful creative flexibility and a slightly more mature, cohesive visual language.
+## Rayflux Horndog
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/RayfluxvX_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/RayfluxvX_00002_.png"
+       alt="Horndog preview"
+       style="max-width:100%; height:auto;">
+</a>
+A model I overtuned on anatomy trying to improve on Photoplus; it tends to produce accidental NSFW content so don't forget to prompt for clothing! High detail, vivid rendering, and strong contrast, but is harder to control than previous models. Suffers a bit from same-face syndrome (use LoRAs to avoid that).
+## Rayflux v3.x → v4
+A consolidated series of refinement experiments designed to progressively “de-hornify” Horndog — cleaner anatomy, fewer unintended biases, and improved stability — while preserving the core Rayflux style. In fine, they offer different flavors than v1-v3 but failed to surpass them.
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_QWEN_IMAGE.png">
+> **Baseline Rayqwest settings (I recommend you experiment):**
+> Sampler: **Euler/Simple** · Steps: **16**
+## RayQwest v1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayqwest_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayqwest_00002_.png"
+       alt="RayQwest preview"
+       style="max-width:100%; height:auto;">
+</a>
+RayQwest v1.0 is designed to remove the characteristic “AI plasticness” often seen in Qwen-Image outputs, restoring a more natural analog grit and the recognizable Ray-style aesthetic. It maintains excellent prompt comprehension, handles a wide range of styles with ease, and runs fast — all while giving Qwen’s already powerful architecture a stronger artistic backbone. Carries some of Qwen’s usual softness — less sharp than Rayflux — but compensates with **excellent prompt accuracy and expressive range**. Works well for portraits, lifestyle photography, stylized art, and creative compositions.
+## RayQwest v0.7
+RayQwest v0.7 is a strong pre-v1 iteration characterized by a heavier analog feel, noticeable grain, and a slightly flatter dynamic range. It’s a bit noisier than v1.0, but compensates with **excellent aesthetic vibes**, maintaining solid prompt comprehension and creative flexibility. Despite being an intermediate version, it remains highly usable and flavorful.
+## RayQwest v0.65
+RayQwest v0.65 was trained on a **different dataset** entirely, giving it a more **dreamlike**, **saturated**, and softer look. It carries a unique atmosphere — less sharp than v0.7 and v1.0, but with an appealing ethereal quality that makes it stand apart. A charming early version that still holds its own artistically.
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_SDXL.png">
+> **Baseline SDXL settings (I recommend you experiment):**
+> Sampler: **DPMPP2M/Karras** · cfg:**5-7** · Steps: **20-30**
+## Raymnants 3.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raymnants_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raymnants_00002_.png"
+       alt="Raymnants preview"
+       style="max-width:100%; height:auto;">
+</a>
+Realistic model with a good sense for Art. Great at portraits, wide ethnic and body type range. The model that started it all. What would happen if I trained a bunch of street photos into an image model? The answer was: RAYMNANTS The rudimentary methodology and complete lack of insight make it nothing short of a miracle that it came out that good. Still took half a dozen iterations before 3.0 though!
+## Rayctifier 1.5
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayctifier_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayctifier_00002_.png"
+       alt="Rayctifier preview"
+       style="max-width:100%; height:auto;">
+</a>
+My artistic model, geared towards painting style and all around illustrations, with a classical painting flair.
+I’ve finetuned a branch of my realistic model, RAYMNANTS, on about 600 public domain paintings from the Smithsonian online archive, concentrating on the 1750-1850 period.
+The result is a model that absorbed some of the style from those classical painting while retaining above average photo capacity. It will perform really well for any prompt destined to output a painting, do well on illustration, but generally underperforms in photo-style context
+## Rayburn 1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayburn_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayburn_00002_.png"
+       alt="Rayburn preview"
+       style="max-width:100%; height:auto;">
+</a>
+My overly opinionated stylized model, geared comic-style illustrations, it will take everything you thrown at it, and give it back, with a twist. I’ve finetuned a branch of my painterly model, RAYCTIFIER, on a mix of about 800 hand-drawn illustrations and synthetic data.
+Careful cross-breeding to cultivate some style traits resulted in a very bold style you can’t really prompt out of but that has a lot of personality. It’s seriously overfit, but that’s what makes it the way it is. The model will perform well at taking its own stylistic stance to any kind of prompt you’ll thrown at it. It’s great fun, and also a bit wild!
+## Rayctifier 1.0
+Rayctifier 1.0 was an earlier version of the model that I thought wasn't deep enough into the painting style I wanted for it. But if you personally find Rayctifier 1.5 is a bit too much, then 1.0 might be just your thing.
+## Raymnants 3.0 LightningW 4L1.0 & 4L2.0
+Lightning distilled variants from Raymnants 3.0, they aimed at accelerating sampling while preserving as much of the original style as possible. YMMV!
+Both versions went through a different distillation process.
+- L1 is closer to the 3.0 look but thinner in detail and nuance.
+> Sampler: **Euler/sgm_uniform** · cfg:**1** · Steps: **8**
+- L2 is more aggressively distilled; punchier but also more chaotic and further from base 3.0.
+> Sampler: **DPMPPSDE/sgm_uniform** · cfg:**2** · Steps: **6**
+## Raymnants 3.4 → 6.2 — Iterative Evolution Line
+Raymnants 3.4 through 6.2 form an exploratory lineage built on 3.0. At the time, the training process involved a lot of experimentation, resulting in stylistic “yoyo effects” between versions — sometimes more realistic, sometimes more expressive, sometimes more saturated.
+At the end of this fluctuating path, the 6.x models stand out as flexible, bright, saturated, and artistically dynamic, if slightly less realistic than 3.0.
+## RayburnBlitz 1.0
+RayburnBlitz 1.0 is the distilled, lightning-fast variant of Rayburn. It remains stylized and comic-driven but loses some of the original model’s extreme boldness that was achieved through overfitting and lost through distillation.
+> Sampler: **Gradient_Estimation/beta57** · cfg:**1.5** · Steps: **6**
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_Z-IMAGE-TURBO.png">
+> **Baseline Rayzist settings (I recommend you experiment):**
+> Sampler: **Euler/Simple** · cfg: **1** · Steps: **8**
+## Rayzist v1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayzist_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayzist_00002_.png"
+       alt="Rayzist preview"
+       style="max-width:100%; height:auto;">
+</a>
+Rayzist v1.0 is a fast, generalist Z-Image Turbo model with unusually strong prompt adhesion for its class. Despite its distilled nature, it's mostly uncensored, has a broad conceptual range, performs well across photography, anime, UI and comic-style rendering. It is also ridiculously good at generating readable text inside the image, which is unheard of for Turbo-tier models. Rapid ideation, multi-style image generation, but 8GB VRAM workflows, photography and anime rendering, and text-in-image generation: Rayzist has it all. Well, almost: it is still a distilled Turbo model and some visual concepts simply aren’t well encoded (e.g., cyclops / one-eyed characters). These gaps are unpredictable and usually only discovered through experimentation.
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_CHROMA.png">
+> **Baseline Raydiant settings (I recommend you experiment):**
+> Sampler: **Euler/Beta** · cfg:**4.5** · Steps: **16**
+> **Notes:** Pad the T5 tokenizer (min_length = 3); works well with dedicated negative prompting.
+## Raydiant v1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raydiant_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Raydiant_00002_.png"
+       alt="Raydiant preview"
+       style="max-width:100%; height:auto;">
+</a>
+Raydiant v1.0 is a Chroma-based model spun off from **Chroma v48**, tuned to produce uncensored analog-style photography with strong contrast, light fuzziness, and a natural film-grain feel. Despite its photographic backbone, it also performs surprisingly well across bold illustrated genres, maintaining expressive color and stylistic weight. It's quite hard to use, with islands of stability in the concepts where it performs well and other places where it's just falling flat on its face. I struggled a lot to train the model and never went back after Chroma v48, but I still like Raydiant for all its tantrums: bleeding concepts, garbled latents and unpolished hands/feet.
+---
+<img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/HF_FLUX-KREA.png">
+> **Baseline Rayflux Krea settings (I recommend you experiment):**
+> Sampler: **ER_SDE/Beta** · Guidance: **3** · Steps: **20** · min/max Shift: **0.5/2**
+## Rayflux Krea v1.0
+<a href="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayflux-Krea_00001_.png" target="_blank">
+  <img src="https://huggingface.co/MutantSparrow/Ray/resolve/main/_IMG_/Rayflux-Krea_00002_.png"
+       alt="Rayflux-Krea preview"
+       style="max-width:100%; height:auto;">
+</a>
+Rayflux Krea v1.0 is a photo style checkpoint tuned for **dreamy, stylish pictures** and a bold flavorful look. It carries a distinct artistic flair and handles a wide variety of styles well, from creative photography to expressive illustration. Compared to standard Rayflux models, it has its own flavor — visually more opinionated than vanilla Flux, but also slightly more resistant to absorbing new concepts. It does suffer a bit from same-face syndrome though, and have some holes in the concepts it covers.
+---
+## Repository Structure
+```text
+KREA2_TURBO/
+    Raykreate_v3.1_INT8_CONVROT.safetensors
+    Raykreate_v1.0.safetensors
+CHROMA/
+    Raydiant v1.0
+FLUX/
+    Rayflux Horndog
+    Rayflux Photoplus
+    Rayflux v1.0
+    Rayflux v1.0 (UNet-only)
+    Rayflux v3.0
+    Rayflux v3.x (3.1–3.4)
+    Rayflux v4.0
+FLUX_KREA/
+    Rayflux Krea v1.0
+QWEN_IMAGE/
+    RayQwest v0.65
+    RayQwest v0.7
+    RayQwest v1.0
+SDXL/
+    Rayburn 1.0
+    RayburnBlitz 1.0
+    Rayctifier 1.0
+    Rayctifier 1.5
+    Raymnants 3.0
+    Raymnants 3.0 LightningW 4L1.0
+    Raymnants 3.0 LightningW 4L2.0
+    Raymnants 3.4
+    Raymnants 3.4b
+    Raymnants 4.0
+    Raymnants 4.1
+    Raymnants 5.0
+    Raymnants 6.0
+    Raymnants 6.1
+    Raymnants 6.2
+Z-IMAGE/
+    Rayzist v1.0
+_IMG_/
+   (all reference images go here)
+_WORKFLOWS/
+   (added ComfyUI workflows references)
+```
+---
+## License
+All models in this zoo are trained on personal and/or mixed-source data and **inherit the license and usage constraints of their respective base models** (see metadata above). Check the upstream repositories for detailed license terms.
+---
+## Credits
+All models, datasets, training, and curation by **Ray / MutantSparrow**.