Wan 2.1 T2V with LyCORIS

Model

Prompt

Negative Prompt

Adapter Repo (e.g., markury/wan-st)

Adapter Path in Repo

Specify for repos with multiple .safetensors files, e.g.: adapter_model.safetensors, pytorch_lora_weights.safetensors, etc.

Adapter Scale

0 2

Scheduler

Flow Shift

2.0-5.0 for smaller videos, 7.0-12.0 for larger videos

1 12

Height

256 1024

Width

256 1792

Number of Frames (4k+1 is recommended, e.g. 33)

17 129

Output FPS

8 30

Guidance Scale (CFG)

1 15

Inference Steps

10 100

Seed (-1 for random)

Set a specific seed for deterministic results

Generated Video

Seed

Tips for best results:

For smaller resolution videos, try lower values of flow shift (2.0-5.0)
For larger resolution videos, try higher values of flow shift (7.0-12.0)
Number of frames should be of the form 4k+1 (e.g., 33, 81)
Stick to lower frame counts. Even at 480p, an 81 frame sequence at 30 steps will nearly time out the request in this ZeroGPU space.