Habr | Project Page | Technical Report | Original GitHub | 🤗 Diffusers

Kandinsky 5.0 I2I Lite Pretrain – Diffusers

Kandinsky 5.0 is a family of diffusion models for video and image generation.

Kandinsky 5.0 Image Edit Lite is a lightweight image-to-image (I2I) generation model with 6B parameters.

The model introduces several key innovations:

Latent diffusion pipeline with Flow Matching for improved training stability
Diffusion Transformer (DiT) as the main generative backbone with cross-attention to text embeddings
Dual text encoding using Qwen2.5-VL and CLIP for comprehensive text understanding
Flux VAE for efficient image encoding and decoding

The original codebase can be found at kandinskylab/Kandinsky-5.

Available Models

Kandinsky 5.0 Image Lite:

model_id	Description	Use Cases
kandinskylab/Kandinsky-5.0-T2I-Lite-sft-Diffusers	6B supervised fine-tuned text-to-image model	Highest generation quality
kandinskylab/Kandinsky-5.0-I2I-Lite-sft-Diffusers	6B supervised fine-tuned image-to-image editing model	Highest generation quality
kandinskylab/Kandinsky-5.0-T2I-Lite-pretrain-Diffusers	6B base pretrained text-to-image model	Research and fine-tuning
kandinskylab/Kandinsky-5.0-I2I-Lite-pretrain-Diffusers	6B base pretrained image-to-image editing model	Research and fine-tuning

Examples

Change the image style to realistic, draw the horse in the same pose, with a black mane and a giant snowball on its nose

Draw this elephant in the same hat inside a green tractor, driving across a yellow field.

Поменяй чебурашку на крокодила Гену из мультфильма, позу и выражения лица оставь прежними

Change the image style to an oil painting; don't change the pose of the animals, but add pronounced oil painting elements - paint strokes

Kandinsky5I2IPipeline Usage Example

import torch
from diffusers import Kandinsky5I2IPipeline
from diffusers.utils import load_image 
# Load the pipeline
model_id = "kandinskylab/Kandinsky-5.0-I2I-Lite-sft-Diffusers"
pipe = Kandinsky5I2IPipeline.from_pretrained(model_id)

_ = pipe.to(device='cuda',dtype=torch.bfloat16)
pipe.enable_model_cpu_offload()                                               # <--- Enable CPU offloading for single GPU inference

# Edit the input image
image = load_image(
    "https://huggingface.co/kandinsky-community/kandinsky-3/resolve/main/assets/title.jpg?download=true"
)

prompt = "Change the background from a winter night scene to a bright summer day. Place the character on a sandy beach with clear blue sky, soft sunlight, and gentle waves in the distance. Replace the winter clothing with a light short-sleeved T-shirt (in soft pastel colors) and casual shorts. Ensure the character’s fur reflects warm daylight instead of cold winter tones. Add small beach details such as seashells, footprints in the sand, and a few scattered beach toys nearby. Keep the oranges in the scene, but place them naturally on the sand."
negative_prompt = ""

output = pipe(
    image=image,
    prompt=prompt,
    negative_prompt=negative_prompt,
    guidance_scale=3.5,
).image[0]

Results


Side-by-side evaluation of T2I SFT on PartiPrompts with extended prompts	Side-by-side evaluation of I2I SFT on the Flux Kontext benchmark with extended prompts

Citation

@misc{arkhipkin2026kandinsky50familyfoundation,
      title={Kandinsky 5.0: A Family of Foundation Models for Image and Video Generation}, 
      author={Vladimir Arkhipkin and Vladimir Korviakov and Nikolai Gerasimenko and Denis Parkhomenko and Viacheslav Vasilev and Alexey Letunovskiy and Nikolai Vaulin and Maria Kovaleva and Ivan Kirillov and Lev Novitskiy and Denis Koposov and Nikita Kiselev and Alexander Varlamov and Dmitrii Mikhailov and Vladimir Polovnikov and Andrey Shutkin and Julia Agafonova and Ilya Vasiliev and Anastasiia Kargapoltseva and Anna Dmitrienko and Anastasia Maltseva and Anna Averchenkova and Olga Kim and Tatiana Nikulina and Denis Dimitrov},
      year={2026},
      eprint={2511.14993},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.14993}, 
}