--- license: apache-2.0 prior: - kandinsky-community/kandinsky-2-1-prior tags: - kandinsky - image-to-image duplicated_from: kandinsky-community/kandinsky-2-1 pipeline_tag: image-to-image --- # Kandinsky 2.1 Kandinsky 2.1 inherits best practices from Dall-E 2 and Latent diffusion while introducing some new ideas. It uses the CLIP model as a text and image encoder, and diffusion image prior (mapping) between latent spaces of CLIP modalities. This approach increases the visual performance of the model and unveils new horizons in blending images and text-guided image manipulation. The Kandinsky model is created by [Arseniy Shakhmatov](https://github.com/cene555), [Anton Razzhigaev](https://github.com/razzant), [Aleksandr Nikolich](https://github.com/AlexWortega), [Igor Pavlov](https://github.com/boomb0om), [Andrey Kuznetsov](https://github.com/kuznetsoffandrey) and [Denis Dimitrov](https://github.com/denndimitrov) ## Usage Kandinsky 2.1 is available in diffusers! ```python pip install diffusers transformers ``` ### Text to image ```python from diffusers import KandinskyPipeline, KandinskyPriorPipeline import torch pipe_prior = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16) pipe_prior.to("cuda") prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting" negative_prompt = "low quality, bad quality" image_emb = pipe_prior( prompt, guidance_scale=1.0, num_inference_steps=25, generator=generator, negative_prompt=negative_prompt ).images zero_image_emb = pipe_prior( negative_prompt, guidance_scale=1.0, num_inference_steps=25, generator=generator, negative_prompt=negative_prompt ).images pipe = KandinskyPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16) pipe.to("cuda") images = pipe( prompt, image_embeds=image_emb, negative_image_embeds=zero_image_emb, num_images_per_prompt=2, height=768, width=768, num_inference_steps=100, guidance_scale=4.0, generator=generator, ).images[0] image.save("./cheeseburger_monster.png") ``` ![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/cheeseburger.png) ### Text Guided Image-to-Image Generation ```python from diffusers import KandinskyImg2ImgPipeline, KandinskyPriorPipeline import torch from PIL import Image import requests from io import BytesIO url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg" response = requests.get(url) original_image = Image.open(BytesIO(response.content)).convert("RGB") original_image = original_image.resize((768, 512)) # create prior pipe_prior = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16) pipe_prior.to("cuda") # create img2img pipeline pipe = KandinskyImg2ImgPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16) pipe.to("cuda") prompt = "A fantasy landscape, Cinematic lighting" negative_prompt = "low quality, bad quality" image_emb = pipe_prior( prompt, guidance_scale=4.0, num_inference_steps=25, generator=generator, negative_prompt=negative_prompt ).images zero_image_emb = pipe_prior( negative_prompt, guidance_scale=4.0, num_inference_steps=25, generator=generator, negative_prompt=negative_prompt ).images out = pipe( prompt, image=original_image, image_embeds=image_emb, negative_image_embeds=zero_image_emb, height=768, width=768, num_inference_steps=500, strength=0.3, ) out.images[0].save("fantasy_land.png") ``` ![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/img2img_fantasyland.png) ### Interpolate ```python from diffusers import KandinskyPriorPipeline, KandinskyPipeline from diffusers.utils import load_image import PIL import torch from torchvision import transforms pipe_prior = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16) pipe_prior.to("cuda") img1 = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" "/kandinsky/cat.png" ) img2 = load_image( "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" "/kandinsky/starry_night.jpeg" ) images_texts = ["a cat", img1, img2] weights = [0.3, 0.3, 0.4] image_emb, zero_image_emb = pipe_prior.interpolate(images_texts, weights) pipe = KandinskyPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16) pipe.to("cuda") image = pipe( "", image_embeds=image_emb, negative_image_embeds=zero_image_emb, height=768, width=768, num_inference_steps=150 ).images[0] image.save("starry_cat.png") ``` ![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/starry_cat.png) ## Model Architecture ### Overview Kandinsky 2.1 is a text-conditional diffusion model based on unCLIP and latent diffusion, composed of a transformer-based image prior model, a unet diffusion model, and a decoder. The model architectures are illustrated in the figure below - the chart on the left describes the process to train the image prior model, the figure in the center is the text-to-image generation process, and the figure on the right is image interpolation.

Specifically, the image prior model was trained on CLIP text and image embeddings generated with a pre-trained [mCLIP model](https://huggingface.co/M-CLIP/XLM-Roberta-Large-Vit-L-14). The trained image prior model is then used to generate mCLIP image embeddings for input text prompts. Both the input text prompts and its mCLIP image embeddings are used in the diffusion process. A [MoVQGAN](https://openreview.net/forum?id=Qb-AoSw4Jnm) model acts as the final block of the model, which decodes the latent representation into an actual image. ### Details The image prior training of the model was performed on the [LAION Improved Aesthetics dataset](https://huggingface.co/datasets/bhargavsdesai/laion_improved_aesthetics_6.5plus_with_images), and then fine-tuning was performed on the [LAION HighRes data](https://huggingface.co/datasets/laion/laion-high-resolution). The main Text2Image diffusion model was trained on the basis of 170M text-image pairs from the [LAION HighRes dataset](https://huggingface.co/datasets/laion/laion-high-resolution) (an important condition was the presence of images with a resolution of at least 768x768). The use of 170M pairs is due to the fact that we kept the UNet diffusion block from Kandinsky 2.0, which allowed us not to train it from scratch. Further, at the stage of fine-tuning, a dataset of 2M very high-quality high-resolution images with descriptions (COYO, anime, landmarks_russia, and a number of others) was used separately collected from open sources. ### Evaluation We quantitatively measure the performance of Kandinsky 2.1 on the COCO_30k dataset, in zero-shot mode. The table below presents FID. FID metric values ​​for generative models on COCO_30k | | FID (30k)| |:------|----:| | eDiff-I (2022) | 6.95 | | Image (2022) | 7.27 | | Kandinsky 2.1 (2023) | 8.21| | Stable Diffusion 2.1 (2022) | 8.59 | | GigaGAN, 512x512 (2023) | 9.09 | | DALL-E 2 (2022) | 10.39 | | GLIDE (2022) | 12.24 | | Kandinsky 1.0 (2022) | 15.40 | | DALL-E (2021) | 17.89 | | Kandinsky 2.0 (2022) | 20.00 | | GLIGEN (2022) | 21.04 | For more information, please refer to the upcoming technical report. ## BibTex If you find this repository useful in your research, please cite: ``` @misc{kandinsky 2.1, title = {kandinsky 2.1}, author = {Arseniy Shakhmatov, Anton Razzhigaev, Aleksandr Nikolich, Vladimir Arkhipkin, Igor Pavlov, Andrey Kuznetsov, Denis Dimitrov}, year = {2023}, howpublished = {}, } ```