bad result on a800

[523997931]

hello,
I use the example code to run the example case. but I found the result is very bad, the second stage result is strange, don't like the project you showed, I don't know where is wrong,I think the first stage is normal, but stage 2 seem has something wrong.

1_0_seed_1_joint.mp4

[jshilong]

Hello, this result does seem unreasonable, as the entire image is filled with artifacts. Did you change the resolution or adjust any parameters in stage 2? It is possible that you are submitting the wrong text prompt to the second stage？(Because it can be seen from the hands and face that the model is trying to generate other things)

Did you run inference on all the prompts? Are none of the prompts able to produce normal results?

[jshilong]

@linjing7
In the #4 , you mentioned that you could run the code and get normal results. Did you have similar strange results at that time?

[linjing7]

No, I got quite good results. Here are some videos I got using prompts from VBench.

2_0_seed_2_joint.mp4.zip

[jshilong]

No, I got quite good results. Here are some videos I got using prompts from VBench.

2_0_seed_2_joint.mp4.zip

Thank you for your feedback. Let me help troubleshoot the issue.

[523997931]

Hello, this result does seem unreasonable, as the entire image is filled with artifacts. Did you change the resolution or adjust any parameters in stage 2? It is possible that you are submitting the wrong text prompt to the second stage？(Because it can be seen from the hands and face that the model is trying to generate other things)

Did you run inference on all the prompts? Are none of the prompts able to produce normal results?

Well, I don't change anything about the config, just run the inf_270_to_1080.sh scripts, just change the node number, so I think it is strange, the resolution is [1920,1080], and the prompt is normal give to the second stage, because I don't change any code ...

[523997931]

2_0_seed_2_joint.mp4.zip

I think it is also strange....for the face part

[jshilong]

Hello, this result does seem unreasonable, as the entire image is filled with artifacts. Did you change the resolution or adjust any parameters in stage 2? It is possible that you are submitting the wrong text prompt to the second stage？(Because it can be seen from the hands and face that the model is trying to generate other things)
Did you run inference on all the prompts? Are none of the prompts able to produce normal results?

Well, I don't change anything about the config, just run the inf_270_to_1080.sh scripts, just change the node number, so I think it is strange, the resolution is [1920,1080], and the prompt is normal give to the second stage, because I don't change any code ...

@523997931 The results of linjing7 are within a reasonable range，the face is related to word description

but in your case I think there is indeed something wrong, because there are no normal results on the hands and face at all.

[jshilong]

Could you share the results for other prompts?
Alternatively, you could run the demo.ipynb and share the outputs.

[523997931]

Could you share the results for other prompts? Alternatively, you could run the demo.ipynb and share the outputs.

result.zip
here is the result, I just run 4, I found it broken, so I stop it, I think the part without person is fine, but for human part, it is really strange

[jshilong]

Now I can confirm that all your results are abnormal, because there seems to be no improvement in image quality.

[523997931]

Now I can confirm that all the results are abnormal, because there seems to be no improvement in image quality.

so what may cause this? I recheck the code, I think the prompt is correct pass to stage2

[jshilong]

Now I can confirm that all the results are abnormal, because there seems to be no improvement in image quality.

so what may cause this? I recheck the code, I think the prompt is correct pass to stage2

I will download this repo and ckpt rerun the code to make sure the latest version is correct

[523997931]

is this normal? first stage num_frames is 13? second is 49?