Provisioning Script and Simple Test to Work the MovieGen Prompts Using FastWan
This project is intended to facilitate easily firing off test renders using the FastWan model across a range of GPU capacities. It uses demo prompts from MovieGenBench and Wan2.2-Lightning on the madiator2011/better-comfyui:slim-5090 container image (aka comfyslim template on Runpod).
-
Start the container, wait for it to fully deploy (wait for it to start Comfy), and open a shell using
zasper/ssh/exec -it bash/whatever. -
Choose the provisioning script that matches your GPU's VRAM and run the corresponding command.
-
For 16GB+ GPUs (Original FP16 Model):
curl -s https://raw.githubusercontent.com/FNGarvin/fastwan-moviegen/main/provision.sh | bash -
For 10GB/12GB GPUs (Quantized GGUF Q6 Model):
curl -s https://raw.githubusercontent.com/FNGarvin/fastwan-moviegen/main/provision10GB.sh | bash -
For 8GB GPUs (Quantized GGUF Q3 Model):
curl -s https://raw.githubusercontent.com/FNGarvin/fastwan-moviegen/main/provision8GB.sh | bash
-
-
Open ComfyUI and load the workflow that corresponds to your setup (found in the "Load Workflow" menu).
FastWanMovieGen.json: Queues a large batch of demo prompts using the full FP16 model. Recommend you also open the FastWan-Simple workflow so you can better track the process of working through the queue.FastWan-Simple.json: A simple template for testing individual prompts with the full FP16 model.FastWan-GGUF.json: A template for low-VRAM GPUs. Important: After loading, you must click the GGUF Loader node and select the.ggufmodel file your provisioning script downloaded from the dropdown menu.
- Custom Prompts: The default prompts are sourced from MovieGenBenchmark and Wan2.2-Lightning, but you can use any text file with one prompt per line by providing the filename.
- Resume Feature: If you stop the batch and later attempt to resume, the script will scan the output directory and attempt to pick up where you stopped.
- Concatenation: If you check the "concat only" option, the script will concatenate all batched videos with nicely labeled chapters that indicate the prompt used for each segment.
This project is licensed under the Creative Commons Attribution-NonCommercial 4.0 International Public License (CC BY-NC 4.0).
This project uses and packages the following, each provided under their respective licenses:
- Prompts from MovieGenBench: Licensed under the CC BY-NC 4.0 license.
- Prompts from Wan2.2-Lightning: Licensed under the Apache License, Version 2.0.
- Models from Wan2.2: Licensed under the Apache License, Version 2.0.
- Models from FastVideo: Licensed under the Apache License, Version 2.0.
- Node from ComfyUI-GGUF: Licensed under the MIT License.