Video on a modest graphics card: Wan 2.2 5B

<!-- lesson: video-small -->

Lesson "video-small", in the group Video of the lesson list. Steps: 13.

#What you will be able to do

You build a one-model Wan 2.2 graph that makes a five-second 720p clip from a prompt, or from an image.

#What you need

• It goes on from Video from text. 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 ti2v and 5b;
  • for clip_name of Load CLIP: a file whose name has umt5;
  • for vae_name of Load VAE: a file whose name has wan2.2_vae.
• Nodes: Load Diffusion Model, Load CLIP, Load VAE, CLIP Text Encode (Prompt), ModelSamplingSD3, Wan22ImageToVideoLatent, 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)

   Wan 2.2 has a small model, 5B (five billion parameters), made for ordinary graphics cards. It draws 24 fps and takes a prompt, an image, or both.

   What you need:

   • the model file, about 10 GB, in models/diffusion_models;
   • the umt5_xxl text encoder, about 7 GB in its fp8 form, in models/text_encoders;
   • the wan2.2_vae file in models/vae.

   Memory: with ComfyUI moving parts of the model to system memory as needed, a card with 8 GB can run it. Time: a five-second clip takes minutes, not seconds, and the first run is longer because the files are read from disk.

   The graph has one model and one sampler, not the pair of the large 14B models. 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 wan2.2_ti2v_5B file in unet_name. ti2v means it takes text, an image, or both.

   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 "Wan 2.2 5B Video Generation" 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 text encoder

   Add a Load CLIP node. Choose the umt5_xxl file in clip_name and set type to wan.

   The text encoder turns the prompt into what the model reads. The fp8 version of the file takes half the memory of the full one and is what the templates use.

4. Load the VAE of the small model

   Add a Load VAE node and choose the wan2.2_vae file in vae_name.

   The 5B model compresses video more strongly than the 14B models and has its own VAE. The wan_2.1_vae file of the large models gives broken frames here.

5. Write the prompt

   Add two CLIP Text Encode (Prompt) nodes and connect the CLIP output of Load CLIP 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 8.

   Shift decides how much of the work goes to the early, noisy steps where the composition and the motion are laid out. With a small value Wan keeps the composition and the motion worse and the clip comes out sluggish.

   The SD3 in the node's name is left from another model: Wan uses the same node.

7. Set the size and the length

   Add a Wan22ImageToVideoLatent node and connect the output of Load VAE to its vae input. Set width to 1280, height to 704 and length to 121.

   121 frames at 24 fps make five seconds. For Wan the number of frames is of the form 4n + 1: 49, 81, 121. The model was trained at this 720p size; a smaller size is faster and less sharp.

8. Starting from an image, if you like (reading)

   The same node has a start_image input. Connect a Load Image node to it and the clip starts from your image; leave it empty and the clip is drawn from the prompt alone. In the template this node is there and switched off.

   With an image, the prompt describes what happens, not what is in the frame. Keep the proportions of the image close to 1280 × 704.

9. Add the sampler

   Add a KSampler node and connect to it:

   • to model, the output of ModelSamplingSD3;
   • to positive and negative, the two prompts;
   • to latent_image, the output of Wan22ImageToVideoLatent.

10. Set up the sampler

In KSampler set steps to 20, cfg to 5, sampler_name to uni_pc and scheduler to simple.

These are the numbers of the official template. Fewer steps are faster and blurrier; a cfg far above 5 burns the colours.

11. 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.

12. 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.

The 5B model draws 24 fps, the large Wan models 16. With another number the clip runs faster or slower than it was meant to.

13. 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.

If the graphics card runs out of memory, the lesson "When a video does not fit in memory" goes on from this very graph.

#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 ti2v and 5b (capitals, dots and dashes do not matter).
• Step 3, Load the text encoder:
  • a Load CLIP node.
  • clip_name of Load CLIP holds a file whose name has umt5 (capitals, dots and dashes do not matter).
  • type of Load CLIP is wan.
• Step 4, Load the VAE of the small model:
  • a Load VAE node.
  • vae_name of Load VAE holds a file whose name has wan2.2_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 Load CLIP into clip of CLIP Text Encode (Prompt) (prompt).
  • CLIP of Load CLIP 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 5 to 8; Do this for me sets 8.
• Step 7, Set the size and the length:
  • a Wan22ImageToVideoLatent node.
  • VAE of Load VAE into vae of Wan22ImageToVideoLatent.
  • width of Wan22ImageToVideoLatent is from 640 to 1280; Do this for me sets 1280.
  • height of Wan22ImageToVideoLatent is from 352 to 1280; Do this for me sets 704.
  • length of Wan22ImageToVideoLatent is from 49 to 121; Do this for me sets 121.
• Step 9, Add the sampler:
  • a KSampler node.
  • MODEL of ModelSamplingSD3 into model of KSampler (a LoraLoaderModelOnly or EasyCache in between is accepted).
  • CONDITIONING of CLIP Text Encode (Prompt) (prompt) into positive of KSampler.
  • CONDITIONING of CLIP Text Encode (Prompt) (negative prompt) into negative of KSampler.
  • LATENT of Wan22ImageToVideoLatent into latent_image of KSampler.
• Step 10, Set up the sampler:
  • steps of KSampler is from 10 to 30; Do this for me sets 20.
  • cfg of KSampler is from 4 to 6; Do this for me sets 5.
  • sampler_name of KSampler is uni_pc.
  • scheduler of KSampler is simple.
• Step 11, 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 12, 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.

#What next

These lessons go on from this one: When a video does not fit in memory, Video guided by another video: Wan 2.2 Fun Control, Video with sound: LTX-2.5, Video from text: HunyuanVideo 1.5, Video from text: Kandinsky 5 Lite, Smoother motion: more frames between frames, The start picture is the anchor, Load, cut and save video: VideoHelperSuite.

#See also

• Wan 2.2
• Video generation
• Video length and frame rate
• How lessons work