Video from text: Kandinsky 5 Lite

<!-- lesson: video-kandinsky -->

Lesson "video-kandinsky", in the group Video of the lesson list. Steps: 12.

#What you will be able to do

You build a light one-model graph that makes a five-second clip from a prompt.

#What you need

• It goes on from Video on a modest graphics card: Wan 2.2 5B. The card says so; it is a hint, not a lock.
• Model files the lesson checks:
  • for unet_name of Load Diffusion Model: a file whose name has kandinsky5 and t2v;
  • for clip_name1 of DualCLIPLoader: a file whose name has qwen_2.5_vl;
  • for clip_name2 of DualCLIPLoader: a file whose name has clip_l;
  • for vae_name of Load VAE: a file whose name has hunyuan_video_vae.
• Nodes: Load Diffusion Model, DualCLIPLoader, Load VAE, CLIP Text Encode (Prompt), ModelSamplingSD3, Kandinsky5ImageToVideo, KSampler, VAE Decode / VAE Decode (Tiled), Create Video, Save Video.
• The app downloads and installs nothing. The step texts below say which files and folders, and which ComfyUI template names them.

#The steps

The words of each step, as the lesson shows them. Node names are in bold.

1. What it costs, first (reading)

   Kandinsky 5.0 Video Lite by Sber AI is a small video model, 2B, about 4 GB on disk. It reads the prompt with two text encoders and decodes with the VAE of HunyuanVideo.

   What you need: the kandinsky5lite_t2v file in models/diffusion_models; qwen_2.5_vl_7b (about 9 GB in fp8) and clip_l in models/text_encoders; hunyuan_video_vae in models/vae. The model itself fits modest cards; the text encoder is the largest file. 50 steps make the run long: expect many minutes.

   If the canvas already holds another graph, start from an empty one: "File ▸ New project".

2. Load the model

   Add a Load Diffusion Model node and choose the kandinsky5lite_t2v file in unet_name.

   If the file you need is not in the list, your ComfyUI does not have it. The app downloads nothing. The easiest way is to open, in ComfyUI itself, the "Kandinsky 5.0 Video Lite Text to Video" template: it shows which files are missing and where to put them. Then choose "Sync ▸ Sync now" in this app's menu, and the files appear in the lists.

3. Load the two text encoders

   Add a DualCLIPLoader node. In clip_name1 choose qwen_2.5_vl_7b, in clip_name2 choose clip_l, and set type to kandinsky5.

4. Load the VAE

   Add a Load VAE node and choose hunyuan_video_vae in vae_name. Kandinsky 5 uses the VAE of HunyuanVideo 1.0.

5. Write the prompt

   Add two CLIP Text Encode (Prompt) nodes and connect the CLIP output of DualCLIPLoader to the clip input of each. In the text field of one, describe the clip: who or what is in the frame, what happens, how the camera moves. The other is the negative prompt: what must not be there.

   For video the action matters most. "A cat jumps onto a windowsill, the camera follows" works better than "a cat on a windowsill".

6. Set the shift

   Add a ModelSamplingSD3 node and connect the MODEL output of the loader to its model input. Set shift to 5.

7. Size and length

   Add a Kandinsky5ImageToVideo node and connect to it the two prompts and the VAE. Set width to 768, height to 512 and length to 121.

   The Lite file of this lesson is made for five-second clips: 121 frames at 24 fps. Its start_image input stays empty for video from text.

8. Add the sampler

   Add a KSampler node. Its model input takes the output of ModelSamplingSD3; positive, negative and latent_image take the outputs of Kandinsky5ImageToVideo with the same names.

9. Set up the sampler

   Set steps to 50, cfg to 5, sampler_name to euler_ancestral and scheduler to beta, as in the template. With fewer steps the run is faster and the clip rougher.

10. Turn the result into frames

Add a VAE Decode node. Connect the sampler's output to its samples input and the output of Load VAE to its vae input.

The sampler gives the clip in compressed form, the latent. This node unfolds it into frames.

11. Make the clip and save it

Add Create Video and Save Video. The frames from VAE Decode go into the images input of Create Video, and its output into the video input of Save Video. Set fps to 24.

Kandinsky 5 draws 24 fps. With another number the clip runs faster or slower than it was meant to.

12. The graph is ready (reading)

Save the graph with "File ▸ Save workflow for ComfyUI…", open the file in ComfyUI and run it there. This app runs nothing in ComfyUI itself.

#What the lesson checks

A reading step is done when you press Got it, next. A doing step ticks itself off when the canvas holds:

• Step 2, Load the model:
  • a Load Diffusion Model node.
  • unet_name of Load Diffusion Model holds a file whose name has kandinsky5 and t2v (capitals, dots and dashes do not matter).
• Step 3, Load the two text encoders:
  • a DualCLIPLoader node.
  • clip_name1 of DualCLIPLoader holds a file whose name has qwen_2.5_vl (capitals, dots and dashes do not matter).
  • clip_name2 of DualCLIPLoader is not the same file as clip_name1 of DualCLIPLoader.
  • type of DualCLIPLoader is kandinsky5.
  • clip_name2 of DualCLIPLoader holds a file whose name has clip_l (capitals, dots and dashes do not matter).
• Step 4, Load the VAE:
  • a Load VAE node.
  • vae_name of Load VAE holds a file whose name has hunyuan_video_vae (capitals, dots and dashes do not matter).
• Step 5, Write the prompt:
  • a CLIP Text Encode (Prompt) node.
  • a CLIP Text Encode (Prompt) node.
  • CLIP of DualCLIPLoader into clip of CLIP Text Encode (Prompt) (prompt).
  • CLIP of DualCLIPLoader into clip of CLIP Text Encode (Prompt) (negative prompt).
  • text of CLIP Text Encode (Prompt) (prompt) holds at least 3 characters.
• Step 6, Set the shift:
  • a ModelSamplingSD3 node.
  • MODEL of Load Diffusion Model into model of ModelSamplingSD3 (a LoraLoaderModelOnly or EasyCache in between is accepted).
  • shift of ModelSamplingSD3 is from 3 to 8; Do this for me sets 5.
• Step 7, Size and length:
  • a Kandinsky5ImageToVideo node.
  • CONDITIONING of CLIP Text Encode (Prompt) (prompt) into positive of Kandinsky5ImageToVideo.
  • CONDITIONING of CLIP Text Encode (Prompt) (negative prompt) into negative of Kandinsky5ImageToVideo.
  • VAE of Load VAE into vae of Kandinsky5ImageToVideo.
  • width of Kandinsky5ImageToVideo is from 512 to 1280; Do this for me sets 768.
  • height of Kandinsky5ImageToVideo is from 320 to 768; Do this for me sets 512.
  • length of Kandinsky5ImageToVideo is from 49 to 121; Do this for me sets 121.
• Step 8, Add the sampler:
  • a KSampler node.
  • MODEL of ModelSamplingSD3 into model of KSampler (a LoraLoaderModelOnly or EasyCache in between is accepted).
  • positive of Kandinsky5ImageToVideo into positive of KSampler.
  • negative of Kandinsky5ImageToVideo into negative of KSampler.
  • latent of Kandinsky5ImageToVideo into latent_image of KSampler.
• Step 9, Set up the sampler:
  • steps of KSampler is from 20 to 50; Do this for me sets 50.
  • cfg of KSampler is from 3 to 7; Do this for me sets 5.
  • sampler_name of KSampler is euler_ancestral.
  • scheduler of KSampler is beta.
• Step 10, Turn the result into frames:
  • a VAE Decode or VAE Decode (Tiled) node.
  • LATENT of KSampler into samples of VAE Decode or VAE Decode (Tiled).
  • VAE of Load VAE into vae of VAE Decode or VAE Decode (Tiled).
• Step 11, Make the clip and save it:
  • a Create Video node.
  • a Save Video node.
  • IMAGE of VAE Decode or VAE Decode (Tiled) into images of Create Video.
  • VIDEO of Create Video into video of Save Video.
  • fps of Create Video is 24.

#See also

• Video generation
• Model families
• How lessons work