NodeWise Studio
RUDownload for Linux

A longer video in parts

<!-- lesson: anchor-parts -->

Lesson "anchor-parts", in the group Video: the anchor picture of the lesson list. Steps: 12.

Only someone who agrees may be brought to life this way: yourself, a made-up character, or a person who gave permission. Obtain permission for the source material and follow local law. For adult (18+) use, obtain consent and follow local law. See Content rules.

What you will be able to do

Two parts of a clip join through a shared anchor frame into one longer video.

What you need

  • It goes on from Two anchors: first and last frame. The card says so; it is a hint, not a lock.
  • Nodes: ModelSamplingSD3, CLIP Text Encode (Prompt), Load VAE, Load Image, WanFirstLastFrameToVideo, KSampler (Advanced), VAE Decode / VAE Decode (Tiled), Create Video, Batch Images.
  • 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. Before you start (reading)

    Only someone who agrees may be brought to life this way: yourself, a made-up character, or a person who gave permission. Obtain permission for the source material and follow local law. For adult (18+) use, obtain consent and follow local law. Exclude anyone under 18 or who appears underage; label synthetic depictions so they do not mislead others. The lessons teach technique in factual, non-graphic terms.

    A video model makes a few seconds at a time. A longer clip is made in parts, and the frame that ends one part is the anchor that starts the next. This lesson adds a second part to the graph of "Two anchors: first and last frame", as the official template with six key frames does.

    What it costs: each part costs as much as the whole graph before, so two parts take twice the time. The memory stays the same: the parts run one after the other.

  2. The first part is needed

    The canvas must hold the graph of "Two anchors: first and last frame": the two shift nodes, the prompts, Load VAE, the image Load Image at the end of the clip, WanFirstLastFrameToVideo, the two samplers KSampler (Advanced) and KSampler (Advanced), VAE Decode and Create Video. If it is not there, do that lesson first.

  3. Load the next key frame

    Add a Load Image node and choose in image the picture the second part ends with. The last picture of the first part is where the second one starts.

  4. The start of the second part

    Add a second WanFirstLastFrameToVideo node. Connect the prompts and the VAE as for the first; connect Load Image to its start_image and Load Image to its end_image. Set width, height and length the same as in the first.

    The shared picture is the seam: the first part ends on it and the second starts on it, so the joint does not jump.

  5. Two passes for the second part

    Add two more KSampler (Advanced) nodes. The first takes the high_noise model through its shift node and the three outputs of WanFirstLastFrameToVideo; the second takes the low_noise model through its own shift node, the prompts of WanFirstLastFrameToVideo and the output of the first. The models are loaded once and serve both parts.

  6. The same split of steps

    Give the new pair the settings of the first part: add_noise enable and disable, the same steps and cfg, the first stops at the same end_at_step where the second has its start_at_step, return_with_leftover_noise enable and disable.

  7. The same seed

    Set noise_seed of KSampler (Advanced) to the same number as in KSampler (Advanced).

    With the same seed and settings the parts start from the same kind of noise, and colour and texture change less at the seam. A part that went wrong can still be redone with another seed.

  8. Turn the second part into frames

    Add a VAE Decode node. Connect the output of the second new sampler to its samples input and the VAE to its vae input.

  9. Join the parts

    Add a Batch Images node. Connect the frames of VAE Decode to its image1 input and those of VAE Decode to image2. Then connect its output to the images input of Create Video in place of the first part alone.

    The template joins its parts with Batch Images in the same way. The shared key frame appears twice at the seam, which at 16 fps is not seen.

  10. Why colour and face drift from part to part, and what holds them (reading)

  • Each part only sees its two anchors. Anything they do not show (the back of a head, a hand) is invented again in each part, a little differently.
  • What holds: the same reference for every part (key frames made from one picture, or the same character LoRA), the same prompt words for the subject, the same seed and settings.
  • Colour: a colour-match node after decoding brings every part to the colours of one reference frame. The Color Match node of KJNodes does this; the official SCAIL extension template matches each part to the last frame of the one before.
  • Detail: a light detail pass on the frames, or an upscale of the joined clip (the lesson "A larger, sharper video: SeedVR2"), evens out small differences.
  1. Without key frames: the last frame as the next anchor (reading)

When you have no picture for the end of each part, take the last frame of the finished part as the start image of the next one: Image From Batch with batch_index -1 and length 1 picks it. The next part is plain image to video from that frame.

Each handover passes through a decoded frame, so small losses add up: after a few parts the face and the colours drift more than with key frames. Keep the chain short, or save each part with Save Video and join the files afterwards in a video editor.

  1. 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, The first part is needed:
    • a ModelSamplingSD3 node.
    • a ModelSamplingSD3 node.
    • a CLIP Text Encode (Prompt) node.
    • a CLIP Text Encode (Prompt) node.
    • a Load VAE node.
    • a Load Image node.
    • a WanFirstLastFrameToVideo node.
    • a KSampler (Advanced) node.
    • a KSampler (Advanced) node.
    • a VAE Decode or VAE Decode (Tiled) node.
    • a Create Video node.
    • IMAGE of Load Image (last frame of the first part) into end_image of WanFirstLastFrameToVideo (start of the first part) (a Upscale Image or Upscale Image By or ImageFromBatch or ImageResizeKJv2 or Resize Image or GetImageRangeFromBatch or ResizeImageMaskNode in between is accepted).
    • latent of WanFirstLastFrameToVideo (start of the first part) into latent_image of KSampler (Advanced) (first part, first pass).
    • LATENT of KSampler (Advanced) (first part, first pass) into latent_image of KSampler (Advanced) (first part, second pass).
    • LATENT of KSampler (Advanced) (first part, second pass) into samples of VAE Decode or VAE Decode (Tiled) (first part).
  • Step 3, Load the next key frame:
    • a Load Image node.
    • image of Load Image (next key frame) is not the same file as image of Load Image (last frame of the first part).
  • Step 4, The start of the second part:
    • a WanFirstLastFrameToVideo node.
    • CONDITIONING of CLIP Text Encode (Prompt) (prompt) into positive of WanFirstLastFrameToVideo (start of the second part).
    • CONDITIONING of CLIP Text Encode (Prompt) (negative prompt) into negative of WanFirstLastFrameToVideo (start of the second part).
    • VAE of Load VAE into vae of WanFirstLastFrameToVideo (start of the second part).
    • IMAGE of Load Image (last frame of the first part) into start_image of WanFirstLastFrameToVideo (start of the second part) (a Upscale Image or Upscale Image By or ImageFromBatch or ImageResizeKJv2 or Resize Image or GetImageRangeFromBatch or ResizeImageMaskNode in between is accepted).
    • IMAGE of Load Image (next key frame) into end_image of WanFirstLastFrameToVideo (start of the second part) (a Upscale Image or Upscale Image By or ImageFromBatch or ImageResizeKJv2 or Resize Image or GetImageRangeFromBatch or ResizeImageMaskNode in between is accepted).
    • width of WanFirstLastFrameToVideo (start of the second part) equals width of WanFirstLastFrameToVideo (start of the first part).
    • height of WanFirstLastFrameToVideo (start of the second part) equals height of WanFirstLastFrameToVideo (start of the first part).
    • length of WanFirstLastFrameToVideo (start of the second part) equals length of WanFirstLastFrameToVideo (start of the first part).
  • Step 5, Two passes for the second part:
    • a KSampler (Advanced) node.
    • a KSampler (Advanced) node.
    • MODEL of ModelSamplingSD3 (shift of the first pass) into model of KSampler (Advanced) (second part, first pass) (a LoraLoaderModelOnly or EasyCache in between is accepted).
    • positive of WanFirstLastFrameToVideo (start of the second part) into positive of KSampler (Advanced) (second part, first pass).
    • negative of WanFirstLastFrameToVideo (start of the second part) into negative of KSampler (Advanced) (second part, first pass).
    • latent of WanFirstLastFrameToVideo (start of the second part) into latent_image of KSampler (Advanced) (second part, first pass).
    • MODEL of ModelSamplingSD3 (shift of the second pass) into model of KSampler (Advanced) (second part, second pass) (a LoraLoaderModelOnly or EasyCache in between is accepted).
    • positive of WanFirstLastFrameToVideo (start of the second part) into positive of KSampler (Advanced) (second part, second pass).
    • negative of WanFirstLastFrameToVideo (start of the second part) into negative of KSampler (Advanced) (second part, second pass).
    • LATENT of KSampler (Advanced) (second part, first pass) into latent_image of KSampler (Advanced) (second part, second pass).
  • Step 6, The same split of steps:
    • add_noise of KSampler (Advanced) (second part, first pass) is enable.
    • steps of KSampler (Advanced) (second part, first pass) equals steps of KSampler (Advanced) (first part, first pass).
    • cfg of KSampler (Advanced) (second part, first pass) equals cfg of KSampler (Advanced) (first part, first pass).
    • end_at_step of KSampler (Advanced) (second part, first pass) equals end_at_step of KSampler (Advanced) (first part, first pass).
    • return_with_leftover_noise of KSampler (Advanced) (second part, first pass) is enable.
    • add_noise of KSampler (Advanced) (second part, second pass) is disable.
    • steps of KSampler (Advanced) (second part, second pass) equals steps of KSampler (Advanced) (first part, first pass).
    • cfg of KSampler (Advanced) (second part, second pass) equals cfg of KSampler (Advanced) (first part, first pass).
    • start_at_step of KSampler (Advanced) (second part, second pass) equals end_at_step of KSampler (Advanced) (second part, first pass).
    • return_with_leftover_noise of KSampler (Advanced) (second part, second pass) is disable.
  • Step 7, The same seed:
    • noise_seed of KSampler (Advanced) (second part, first pass) equals noise_seed of KSampler (Advanced) (first part, first pass).
  • Step 8, Turn the second part into frames:
    • a VAE Decode or VAE Decode (Tiled) node.
    • LATENT of KSampler (Advanced) (second part, second pass) into samples of VAE Decode or VAE Decode (Tiled) (second part).
    • VAE of Load VAE into vae of VAE Decode or VAE Decode (Tiled) (second part).
  • Step 9, Join the parts:
    • a Batch Images node.
    • IMAGE of VAE Decode or VAE Decode (Tiled) (first part) into image1 of Batch Images.
    • IMAGE of VAE Decode or VAE Decode (Tiled) (second part) into image2 of Batch Images.
    • IMAGE of Batch Images into images of Create Video.

See also

This page is imported from the application repository's wiki (docs/wiki/lessons/Anchor-Parts.md, revision b8e3f71). The same guide is built into the app: Help ▸ User guide.

To refresh it, run npm run import and then npm run build in the website folder.