stage = 'no' issue about storygen

[Verg-Avesta]

Sorry, I just found a bug in pipeline.py and fixed it. Please use the new pipeline.py .
Now when using stage='no' with a random prev_p and ref_image, you can generate the correct result.

[lq-blackcat]

You need to set stage='multi-image-condition' or stage='auto-regressive' for img_conditions to have a non-empty value.

[chwjshlwel]

#10 (comment)
Here they said if want to generate a story, we should set stage='no'

I fact, I want to generate a story like the Fig. 1 in the paper, but I do not know how to do it.
When set stage='multi-image-condition' or stage='auto-regressive', the generated images are so different and not consistent.

[haoningwu3639]

To generate the first frame of the story, you need to set stage = 'no', because there is no existing conditional frame yet.
Later, you can use the generated first frame as a condition to auto-regressive synthesize the story, that is, set stage = 'multi-image-condition' or stage = 'auto-regressively'.
Additionally, as mentioned in previous issues (#10 (comment)). 'descriptive text' is more suitable to serve as a prompt than 'narrative story text'. Thus, we also mentioned in our paper that we should use ChatGPT or GPT-4 (or an open-source LLM such as LLaMA) to convert the storylines into corresponding descriptions as input to the model. We recommend you do the same.

[lq-blackcat]

@haoningwu3639 AttributeError: 'NoneType' object has no attribute 'items'，this does not explain how to solve this problem.

[Verg-Avesta]

Please refer to #10 (comment) .
When using stage='no', you should use a random prev_p and ref_image instead of empty ones. The model will not extract and attend to their features when stage = 'no'.

[EIKO1241]

Hi!I set the stage="no" and set the prev_p=["The yellow-haired man"] and ref_image=["/gemini/code/StoryGen/data/boy.png"] instead of empty ones.But the error that "nonetype" object has no attribute "items" remains.How should I fix it?