Commit
•
d285250
1
Parent(s):
e5364f6
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,4 +1,8 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
LoRA finetunes can be used on the CogVideoX-Fun Pose models as well, this is not just applicable to the InPaint model which this was trained on.
|
| 3 |
|
| 4 |
Here are a few samples generated by the later finetuned model at 6.5k steps (These samples are more trained than the current model upload):
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model:
|
| 3 |
+
- alibaba-pai/CogVideoX-Fun-5b-InP
|
| 4 |
+
---
|
| 5 |
+
This video LoRA model upload is an early finetuned research version based on [CogVideoX-Fun-V1.1-5B-INP](https://huggingface.co/alibaba-pai/CogVideoX-Fun-V1.1-5b-InP) on a dataset of 65 captioned .mp4 video samples at a resolution of 720x480 based on the film Bladerunner 2049 for 2k steps on single H100 gpu using parameters provided by CogvideoX Fun's DeepSpeed finetune script. This finetune is an early version as we prepare to train better versions, in order to inference this currently it should be placed in a directory which can be read by the predict_t2v.py script provided by CogVideoX-Fun, as currently I believe there is no ComfyUI node for LoRAs.
|
| 6 |
LoRA finetunes can be used on the CogVideoX-Fun Pose models as well, this is not just applicable to the InPaint model which this was trained on.
|
| 7 |
|
| 8 |
Here are a few samples generated by the later finetuned model at 6.5k steps (These samples are more trained than the current model upload):
|