Website powered by

LoRA Training (Kohya_ss) – Custom Flux Model via Kohya_ss

πŸ”Ή Project Name: LoRA Identity Training – Kohya + ComfyUI Workflow

πŸ”Ή Overview:
This project demonstrates my full workflow for training a LoRA model from scratch in Kohya_ss, using my own dataset and style objectives β€” with integration into ComfyUI for stylized character generation.

The result is a custom Flux LoRA that retains my facial identity across a wide range of environments and aesthetics.
It serves as a foundation for character consistency in stylized worlds, trailers, or game-ready pipelines.

πŸ”Ή Key Highlights:
– Trained in Kohya_ss with >30 annotated and masked photos
– Tokenized tagging for expression and emotion variation
– Identity remains stable across lighting, outfits, and context
– Output generated via ComfyUI with Flux model integration

πŸ”Ή Visual Examples Include:
– Retro sci-fi traveler
– Fantasy paladin with oil-paint texture
– Digital ghost in Tokyo alley
– Vlogger in VHS cyberworld
– Soft light anime boy
–and more

πŸ”Ή Tech Stack:
Kohya_ss for training, Flux model as base, ComfyUI for generation, manual prompt engineering and dataset prep

πŸ”Ή Use Cases:
– Stylized NPC/hero generation
– AI creator avatars
– Game trailers, virtual influencers
– Consistent identity content

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

I trained a Flux LoRA on my own dataset using Kohya_ss, enabling precise identity generation across multiple styles in ComfyUI.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.

Meet Dexy β€” a stylized AI character trained from scratch using a custom LoRA.
Animated, expressive, and ready to speak for his creator.