ltx video found footages workflow

5.0

0 reviews
6
2.7K
596
0
Description

- You can use the schelduder smg uniform instead of linear quadratic , if you don't use linear quadratic enable the shift sigmas node


- you must have ollama installed in your machine : https://ollama.com/


- If you can use qwen2 vl or minicpm-v2 as vlm , it generates the best prompts which increases the results. minicpm-v1 is deprecated and can sometimes generate bad outputs.


qwen2 vl 7b : https://github.com/IuvenisSapiens/ComfyUI_Qwen2-VL-Instruct


- use ltx version 0.9.1


- to have good results it is advised to use low resolution images with slightly tilted camera angles that look like found footage clips. you don't have to use 512x512 resolution, it can work with larger resolutions.


- good results around 30-50 steps


- I use pinterest to find images of liminal spaces


- if you have any ideas on how to improve the workflow or suggestions you can contact me: https://www.reddit.com/user/Qparadisee/


Have fun!

Node Diagram
Discussion

(No comments yet)

Author

2
1.2K
13
6.6K

Reviews

No reviews yet

Versions (1)

  • - latest (3 months ago)

Primitive Nodes (11)

LTXVApplySTG (1)

LTXVLoader (1)

LTXVModelConfigurator (1)

LTXVShiftSigmas (1)

Note (6)

STGGuider (1)

Custom Nodes (18)

ComfyUI

  • - CLIPTextEncode (2)

  • - KSamplerSelect (1)

  • - BasicScheduler (1)

  • - SamplerCustomAdvanced (1)

  • - VAEDecode (1)

  • - CLIPLoader (1)

  • - LoadImage (1)

  • - RandomNoise (1)

  • - OllamaVision (1)

  • - show_text_party (1)

  • - VHS_VideoCombine (1)

  • - StringConstant (1)

  • - ImageResizeKJ (1)

  • - INTConstant (2)

  • - StringConstantMultiline (1)

  • - String Replace (mtb) (1)

Checkpoints (0)

LoRAs (0)