Kandinsky 5.0 Video Lite Image to Video
A lightweight 2B model that generates videos from English and Russian prompts with high visual quality.
简介
The 'Kandinsky 5.0 Video Lite Image to Video' workflow is designed to transform static images into dynamic video sequences using the Kandinsky model. This workflow leverages a lightweight 2B model capable of generating high-quality videos from prompts in both English and Russian. It uses a series of nodes, including SaveVideo, LoadImage, and MarkdownNote, to streamline the process from image input to video output.
The workflow is particularly optimized for generating videos with dimensions of 768x512, ensuring that the width and height are divisible by 128 for optimal performance in longer sequences. This attention to detail in resolution ensures that the resulting video maintains a high visual quality while remaining computationally efficient.
Technically, this workflow operates by first loading an image using the LoadImage node. The image is then processed through the Kandinsky model, which interprets the visual data and any accompanying text prompts to create a cohesive video narrative. The SaveVideo node is used to compile the generated frames into a video file. This workflow is especially useful for artists and content creators looking to add motion to their static images, providing a seamless method to enhance visual storytelling with minimal setup.
你需要提供
- 正在读取工作流
在 ComfyUI 中运行。 此工作流无法在本页面运行。你可以在 Comfy Cloud 中打开（已预先加载所需内容），或下载后在自己的 ComfyUI 中使用。
在 Cloud 中试用查看工作流内部
节点图的只读预览。在 Comfy Cloud 中打开或下载后即可编辑节点。
Loading the graph
Open full-size workflow preview- 作者
- ComfyUI
需要自定义节点
- 正在读取工作流
通过代码使用此工作流
尽可能根据工作流图和输入自动生成代码，否则显示已保存的示例。运行代码可能产生费用，请先核对凭据和端点。
What you need
- Runs on
- Comfy Cloud
- 密钥
- COMFY_API_KEY
需要付费的 Cloud 方案和额度。
# Python 3.10+: pip install comfy-sdk==0.4.0
import os
from comfy_sdk import Comfy
api_key = os.environ.get("COMFY_API_KEY", "").strip()
if not api_key: raise RuntimeError("Set COMFY_API_KEY")
os.environ.setdefault("COMFY_BASE_URL", "https://cloud.comfy.org")
client = Comfy(api_key=api_key)
workflow = client.workflows.from_json({
"10": {
"class_type": "SaveVideo",
"inputs": {
"filename_prefix": "video/Kandinsky5",
"format": "mp4",
"format.codec": "h264",
"video": [
"69:66",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"11": {
"class_type": "LoadImage",
"inputs": {
"image": "crystal_flower.png"
},
"_meta": {
"title": "Load Image"
}
},
"69:53": {
"class_type": "UNETLoader",
"inputs": {
"unet_name": "kandinsky5lite_i2v_5s.safetensors",
"weight_dtype": "default"
},
"_meta": {
"title": "Load Diffusion Model"
}
},
"69:54": {
"class_type": "VAELoader",
"inputs": {
"vae_name": "hunyuan_video_vae_bf16.safetensors"
},
"_meta": {
"title": "Load VAE"
}
},
"69:55": {
"class_type": "DualCLIPLoader",
"inputs": {
"clip_name1": "qwen_2.5_vl_7b_fp8_scaled.safetensors",
"clip_name2": "clip_l.safetensors",
"device": "default",
"type": "kandinsky5"
},
"_meta": {
"title": "Load CLIP (Dual)"
}
},
"69:56": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"69:55",
0
],
"text": "Cinematic dynamic scene in deep blue dream haze: The camera glides slowly forward, focusing on tightly closed crystal flower buds (starlight-sparkled, translucent) that sway gently with misty currents. As the lens pans right, more closed blooms blur into pulsing bokeh; sequined leaves tremble, scattering glimmers that trail light streaks. The camera then pulls back softly, the blue tone deepening from cyan to indigo, while the buds’ subtle sway syncs with drifting mist—all motion flowing in a quiet, breath-like rhythm, wrapping the scene in immersive, glittering cinematic serenity."
},
"_meta": {
"title": "CLIP Text Encode (Positive Prompt)"
}
},
"69:57": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"69:55",
0
],
"text": ""
},
"_meta": {
"title": "CLIP Text Encode (Negative Prompt)"
}
},
"69:58": {
"class_type": "ImageScale",
"inputs": {
"crop": "center",
"height": [
"69:60",
0
],
"image": [
"11",
0
],
"upscale_method": "lanczos",
"width": [
"69:59",
0
]
},
"_meta": {
"title": "Upscale Image"
}
},
"69:59": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 768
},
"_meta": {
"title": "Width"
}
},
"69:60": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 512
},
"_meta": {
"title": "Height"
}
},
"69:61": {
"class_type": "ModelSamplingSD3",
"inputs": {
"model": [
"69:53",
0
],
"shift": 5
},
"_meta": {
"title": "ModelSamplingSD3"
}
},
"69:62": {
"class_type": "Kandinsky5ImageToVideo",
"inputs": {
"batch_size": 1,
"height": [
"69:60",
0
],
"length": 121,
"negative": [
"69:57",
0
],
"positive": [
"69:56",
0
],
"start_image": [
"69:58",
0
],
"vae": [
"69:54",
0
],
"width": [
"69:59",
0
]
},
"_meta": {
"title": "Kandinsky5ImageToVideo"
}
},
"69:63": {
"class_type": "KSampler",
"inputs": {
"cfg": 5,
"denoise": 1,
"latent_image": [
"69:62",
2
],
"model": [
"69:61",
0
],
"negative": [
"69:62",
1
],
"positive": [
"69:62",
0
],
"sampler_name": "euler_ancestral",
"scheduler": "beta",
"seed": 611409204875324,
"steps": 50
},
"_meta": {
"title": "KSampler"
}
},
"69:64": {
"class_type": "ReplaceVideoLatentFrames",
"inputs": {
"destination": [
"69:63",
0
],
"index": 0,
"source": [
"69:62",
3
]
},
"_meta": {
"title": "Replace Video Latent Frames"
}
},
"69:65": {
"class_type": "VAEDecode",
"inputs": {
"samples": [
"69:69",
0
],
"vae": [
"69:54",
0
]
},
"_meta": {
"title": "VAE Decode"
}
},
"69:66": {
"class_type": "CreateVideo",
"inputs": {
"bit_depth": "auto",
"codec": "none",
"color_space": "sRGB",
"fps": 24,
"images": [
"69:65",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"69:69": {
"class_type": "NormalizeVideoLatentStart",
"inputs": {
"latent": [
"69:64",
0
],
"reference_frame_count": 5,
"start_frame_count": 4
},
"_meta": {
"title": "NormalizeVideoLatentStart"
}
}
})
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)常见问答
For best results, use an image with dimensions of 768x512, ensuring both width and height are divisible by 128.
This workflow is optimized for English and Russian prompts. Other languages may not produce the desired results.
The workflow supports generating videos of 5 or 10 seconds, depending on the configuration settings.
Ensure that your image dimensions are correct and adjust the text prompts or model settings to refine the output.