My stuff

  • My Workflows

  • Liked Workflows

  • Following Workflows

Go to OpenArt main site
Upload workflow

[AVD] Very Fast Img2Vid (+extras)

5.0

0 reviews
9
11.7K
1.7K
1
Description

Workflow for Advanced Visual Design class.


Note: This requires KJNodes (not in comfymanager) for the GET and SET nodes:  https://github.com/kijai/ComfyUI-KJNodes

You can install it in the ComfyManager by going to "Install via Git URL" in the options in the Manager. Otherwise may need to do it manually.


This workflow has some options to it, but is generally an image to video workflow.  


You load in an image, and it is applied as an IPAdapter at full strength, along with a text prompt.  It runs through an efficient loader that uses any SD1.5 model and a specific AnimateLCM LoRA (t2v) that allows it to be optimized for fast sampling using the LCM sampler in the context of video.


AnimateDiff is used to create the video, and there is an option for enabling/disabling (highlight and use CTRL-B) MotionLoRAs to control the movement within the render (keep their strength below .7).


It uses the "SparseCTRL" ControlNet which helps maintain the same image between frames when it is turning the image into a video.  


Those are the basics.


There are also a lot of extra options in this workflow:


  • Prompt Extractor is a node that can extract a prompt from an image. This is useful if you are not getting super accurate results to your image, to push more the details of the original image (or other image if you are experimenting)


    • --->To turn this on, highlight the purple nodes in the initialization section and press CTRL-B. After, you will need to make sure the Switch node is wired to the Set Prompt node. Switching between true and false will allow you to choose between using a written prompt (true) or an extracted prompt (false)


    • -SDXL Lightning Image Generation is a section that allows to generate a new image from the text prompt before running it into the image to video section. You can turn this on in the Fast Groups Bypasser at the top left of the workflow. It is turned off by default


    • -Upscale and 2nd Pass Render is a section that upscales the original frames and then runs a second render pass. This increases the resolution and quality, at the price of speed. Generally you'd want to test all your settings before turning this on, and use this section for when you are ready for more final renders.


    • -Interpolate Frames is a section to smooth out the end video render
Discussion

(No comments yet)

Loading...

Author

5
5.1K
47
29.3K
    image.webp (55.7 kB)

No reviews yet

  • - latest (a year ago)

Primitive Nodes (56)

Fast Groups Bypasser (rgthree) (1)

GetNode (18)

Note (16)

Primitive integer [Crystools] (3)

Reroute (2)

SetNode (13)

Switch any [Crystools] (2)

Switch image [Crystools] (1)

Custom Nodes (41)

AnimateDiff Evolved

  • - ADE_ApplyAnimateDiffModelSimple (1)

  • - ADE_LoopedUniformContextOptions (1)

  • - ADE_AnimateDiffSamplingSettings (1)

  • - ADE_AnimateDiffLoRALoader (3)

  • - ADE_LoadAnimateDiffModel (1)

  • - ADE_UseEvolvedSampling (1)

ComfyUI

  • - UpscaleModelLoader (1)

  • - PreviewImage (1)

  • - CLIPVisionLoader (1)

  • - LatentUpscaleBy (1)

  • - LoadImage (1)

  • - ImageScaleBy (1)

  • - ImageUpscaleWithModel (1)

  • - VAEEncode (1)

  • - ToBasicPipe (1)

  • - FromBasicPipe (1)

  • - PromptExtractor //Inspire (2)

  • - IPAdapterModelLoader (1)

  • - IPAdapterAdvanced (1)

  • - ACN_SparseCtrlSpreadMethodNode (1)

  • - ACN_SparseCtrlLoaderAdvanced (1)

  • - ACN_SparseCtrlRGBPreprocessor (1)

  • - ACN_AdvancedControlNetApply (1)

  • - VHS_VideoCombine (3)

  • - Text Input [Dream] (2)

  • - KSampler (Efficient) (3)

  • - Efficient Loader (2)

  • - SizeSelector (1)

  • - Seed Everywhere (1)

Checkpoints (0)

LoRAs (3)

v2_lora_PanLeft.ckpt

v2_lora_ZoomIn.ckpt