Magic prompt generation produces different results for first run, then consistently produces results for repeated runs

[unnamedplugins]

Describe the bug
When running a seed with a magic prompt, the first gen will produce a different image than subsequent gens. This is visible in x/y grids and manual runs.

To Reproduce
Steps to reproduce the behaviour, include the prompt you used if applicable:

1. Enter a prompt of your choice
2. Set a static seed
3. Enable magic prompt with any settings
4. Generate an image
5. Generate a second image
6. Generate a third image
7. The first image will be very different from the other two images.

Alternative way to reproduce

1. Set the same settings for dynamic prompts as above
2. Use the x/y plot script
3. Set the X axis to "Seed", enter in the same seed at least three times (below I used 1,1,1)
4. The first image will be very different from subsequent gens, and use a different magic prompt
   

Also mention which checkboxes have been set.

Expected behaviour
The first gen should use the same magic prompt text as the subsequent images.

Additional context
This appears to be something to do with the GPT2 pipeline, but I am unable to identify any workarounds.
I tried:

1. Initializing the pipeline every call to the script
2. Pre-running the pipeline with the incoming prompt and throwing the results away (off-by-one)

I first found the issue when doing x/y plots, but determined it could be repeated by manually changing a seed and re-running the same image generation repeatedly. It is able to be reproduced with any batch size and count.

The trigger seems to be closely related to changing the seed.

[adieyal]

Could you please test 0.29.15 and let me know if it's fixed? I couldn't track down the issue so I'm forcefully setting the random seed in the magic prompt initialiser

[unnamedplugins]

That appears to have fixed it as far as I can tell. Thanks!