ComfyUI beginner friendly Low VRAM Kandinsky 5 Lite Image-to-Image Workflow with Easy Prompt Saver by Sarcastic TOFU — Kandinsky_I2I_v1.0

This is a very simple workflow that helps you to save your Kandinsky 5 Lite Image to Image Generation Data into a human readable .txt file. This will automatically get and write your metadata to the .txt file. You will find all the saved prompt files that it generated with the images inside the Archive (.Zip) that has the workflow. Also with the Image Saver Simple node used you may embed the workflow itself with each saved image or save the image and workflow for your work separately.

The Kandinsky-5.0-T2I-Lite model and its I2I variant are major milestones for the Russian AI landscape, having been developed by Sber AI and SberDevices in collaboration with the AIRI (Artificial Intelligence Research Institute). Released in November 2025, this 6-billion-parameter model family was specifically trained to excel at understanding Russian cultural concepts and linguistic nuances while remaining highly efficient for general use. The architecture, which utilizes a Cross-Attention Diffusion Transformer (CrossDiT) and dual text encoders like Qwen2.5-VL, allows it to produce high-resolution imagery at 1K quality with a focus on strong text rendering within the images themselves. This project stands as a central part of Russia's push into open-source generative foundation models, providing a lightweight yet powerful tool for both text-to-image and image-to-image synthesis that rivals much larger global counterparts. From my usage I can say that Kandinsky Image 5 Lite T2I & I2I models fit somewhere between Z-Image Base (or Flux.2 Klein) and very hunky Flux.2 Dev model in it's very unique way. I think this needs more attention, LORA and checkpoint trainings. Even without LORA this can do better unfiltered exposed anatomical details and skin textures (especially for European subjects) than Z-Image or Flux.2. (look at my example images generated with this workflow). I used a GGUF clip file I used once as an alternative second clip file than the one suggested by the developers of this model and even then it performed good with my 8GB VRAM OCULink connected eGPU. I provided more details below if you want to use full non-GGUF second clip file.

You can download your necessary Kandinsky 5 Lite Image-to-Image model used from HuggingFace (Details are mentioned below). Make sure you have latest enough ComfyUI installation and install any necessary nodes for for this workflow using ComfyUI manager and place the correct files in correct places. Also check out my other workflows for SD 1.5 + SDXL 1.0, Pony, WAN 2.1, WAN 2.2, MagicWAN Image v2, QWEN, HunyuanImage-2.1, HiDream, KREA, Chroma, AuraFlow, NoobAI, Illustrious, Lumina2, Z-Image Turbo, Flux.2 Klein and Flux.1 . Feel free to toss some yellow Buzz on stuffs you like.

How to use this -

#1. Just select your Kandinsky 5 Lite Image-to-Image model files first and now

#2. select your desired reference image to start

#3. then input your desired image to image prompt.

#4. select how many images you want (Change the number besides the "Run" button)

#5. select image sampling methods, CFG, steps etc. settings

#6. finally press the run button to generate. That's it..

** Please note that Kandinsky 5 Lite Text-to-Image model and Kandinsky 5 Lite Image-to-Image model are two completely separate models and if you use one for other's task your outputs will not be as expected.

Required Files

===============

Download Link for Kandinsky 5 Lite Image-to-Image Model -

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

https://huggingface.co/kandinskylab/Kandinsky-5.0-I2I-Lite/resolve/main/model/kandinsky5lite_i2i.safetensors

Download Link for Kandinsky Image 5 Lite Text Encoders -

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Flux.1 Clip L Text Encoder (This is same as Flux.1, if you already have one you don't need to download again) -

---

https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors

Dumpling-Qwen2.5-VL-7B-GGUF Text Encoder (One used in the workflow) -

---

https://huggingface.co/mradermacher/Dumpling-Qwen2.5-VL-7B-GGUF/resolve/main/Dumpling-Qwen2.5-VL-7B.Q2_K.gguf

**alternatively you can use any one among these very big non-GGUF Qwen2.5 VL 7B text encoders from HunyuanVideo_1.5_repackaged repo ( https://huggingface.co/Comfy-Org/HunyuanVideo_1.5_repackaged/tree/main/split_files/text_encoders ) as your second clip file if you have very hunky GPU with 16GB or 24GB or 32GB GPU. See which one fits on your system.

Download Link for Kandinsky Image 5 Lite VAE -

++++++++++++++++++++++++++++++++++++++++++++++++++

(This is same as Z-Image VAE, if you already have one you don't need to download again) -

---

https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors

**alternatively you can use the UltraFlux VAE if you want from my CivitAI profile ( https://civitai.com/models/2245573/ultraflux-vae-mirrored-from-hugging-face-repo ) or other link mentioned on that post.