Model Description

This model is designed to eliminate the "oily/glossy skin" texture often found in Illustrious checkpoints, delivering a Furry Bara checkpoint rich in artistic flair. You can expect images with exceptional lighting and distinct stylistic qualities.

Merge Process / Recipe

Base Construction: Performed a layered merge of Indigo_Furry_Mix_06 and ShadowForge to create the core Model 1.
Style Extraction: Integrated style LoRAs into Red Lily, then subtracted Illustrious 0.1 to extract a pure Style-Oriented LoRA.
Final Assembly: The Style-Oriented LoRA was layer-pruned and merged into Model 1 alongside Kombagavad_Gita3.9 and NAI_vpred_fix at specific ratios.

Update Notes

⚠️ Important Note: Each version in this repository is designed with a distinct focus. A higher version number does not necessarily mean it is "better" or the right fit for your specific needs. Please choose the version that best aligns with your creative goals.

Version 4.0

In this version, I have significantly adjusted the merging formula and shifted the foundational base model to cknoob (Chenkin Noob XL ckxl).

The primary objective of this update is to drastically weaken the heavy baked-in base style of previous versions. This makes the model much more versatile, flexible, and responsive to your specific prompts and LoRAs. However, as an unavoidable trade-off of this generalization, you may occasionally encounter the smoother, somewhat "oily" or glossy skin textures commonly seen in other standard models.

Merge Process / Recipe:

Style Extraction: Extracted a difference LoRA from ShadowForge by pruning it against Noob 1.1.
Base Injection: Merged this newly extracted ShadowForge LoRA into cknoob 0.2.
Final Assembly: Performed a precise layered block-merge between the resulting model and Indigo 11.

Version 1.5 (Generalization & Compatibility Focus)

Built upon the foundation of v1.0, this version intentionally dials back the heavy stylization to offer a much more versatile and generalized base. The generation experience feels closer to traditional foundational models, providing significantly improved compatibility with other community resources (like your favorite style LoRAs). Additionally, v1.5 excels at handling complex compositions, performing well with negative space, wide-angle perspectives, and panoramic shots.

Version 3.0 (Concept & Character Focus)

This version doubles down on Furry-related concepts, pushing the boundaries of character anatomy while actively restructuring the underlying stylistic composition. Because of its intense focus on the subject, v3.0 is recommended for artworks where the character is the absolute focal point and occupies a large portion of the frame (such as portraits, close-ups, or dynamic character-driven scenes).

Version 2.1

Anatomy Optimization: Significantly improved anatomical correctness for more stable and natural posing.
Enhanced Furry Concept: Strengthened the model's understanding of furry aesthetics and specific traits.
Higher Success Rate: Increased the "good image" ratio, making it easier to get high-quality results with fewer prompts.
Style Shift: Please note that these improvements have led to a slight evolution in the base artistic style compared to v1.0.

Regarding Version 1.0 (Refined) If you still prefer the aesthetic of the original version, I have re-uploaded a fixed v1.0 (TE Fix).

TE Fix: I repaired the Text Encoder, which was likely the cause of previous high error rates and poor LoRA compatibility.
The core functionality remains the same, but it should now be much more stable.

⚠️ 1.0 Version Important Note:

Due to the heavy stylization inherent to this model, you may occasionally encounter anatomical irregularities, such as missing limbs. The model is also quite sensitive to prompting. For the best user experience and significantly improved stability, I highly recommend pairing this checkpoint with your favorite Style LoRAs.

Merge Process / Recipe

Base Construction: Performed a layered merge of Indigo_Furry_Mix_06 and ShadowForge to create the core Model 1.
Style Extraction: Integrated style LoRAs into Red Lily, then subtracted Illustrious 0.1 to extract a pure Style-Oriented LoRA.
Final Assembly: The Style-Oriented LoRA was layer-pruned and merged into Model 1 alongside Kombagavad_Gita3.9 and NAI_vpred_fix at specific ratios.

Recommended Settings

Quality Tags:

Positive: masterpiece, very aesthetic, best quality, very awa, amazing quality, absurdres, newest, high detail
Negative: worst quality, worst detail, worst aesthetic, flat color, jpeg artifacts, distortion, ai-generated, scan artifacts, malformed hands, cluttered details, noisy texture, watermark, signature, text, female, girly, bad anatomy, bad proportions, ugly, fusion, mutated, mutation, deformed, distorted, (missing limb:1.2), (disconnected limbs:1.1), (malformed limbs:1.1), flexible deformity, disfigured, fused eyes, skinny, broken tail, bad tail

Lighting & Style Tags:

chiaroscuro, Natural Light, Mood Lighting, ray tracing, dappled light
Default Style Enhancers: cinematic lighting, volumetric lighting, dramatic shadows, impressionism, blurred background, atmospheric, depth of field, dynamic angle

Generation Parameters:

CFG Scale: 3.5 - 7
Sampler: Euler a + SGM Uniform OR DPM++ 2M + Karras
Steps: 22+
Upscaler: 4x-UltraSharp or 001_classicalSR_DF2K_s64w8_SwinIR-M_x4

Silent_Era-FurryMix-NAI_XL | 沉默年代 - V5.1 （stylized V4）