Do not work with SD.Next Diffusers

[kahsolt]

Do not work with SD.Next Diffusers

[Kahsolt]

hello~ where can I find this fork "SD.Next"? it seems alike a new front end...?
I'd like to check it and decide whether to support it in the master or a new branch of this repo :)

[Kahsolt]

Thx~ I'll check it soon :)

[Kahsolt]

thx for the info, I've just installed the vlad's and started to test.
you may be right, I guess that hooks for the original backend might not work for diffuser backend
I'm inspecting deeper

[sebaxakerhtc]

This DictWithShape is added in A1111's stream to support SDXL

On Vlad's SD.Next there's no needs for this, so - maybe we need to solve just processing_diffusers, imho
But i'm not a dev - you know better :)

[sebaxakerhtc]

hello~ where can I find this fork "SD.Next"? it seems alike a new front end...? I'd like to check it and decide whether to support it in the master or a new branch of this repo :)

Hey, it's just a number 1 fork (vladmandic - automatic) I forget it was renamed :) sorry

https://github.com/vladmandic/automatic

[sebaxakerhtc]

Thank you!
If you'll need any info - let me know!

[Kahsolt]

Here's a question~
You are NOT using a SDXL model ckpt right?

This DictWithShape is added in A1111's stream to support SDXL, and the vlad's may come up with it later (?)
You can try revert ptravel version to v3.0 (commit 98a6523)

cd extensions/stable-diffusion-webui-prompt-travel
git checkout 98a6523

and check if it works for you temporarily

[sebaxakerhtc]

You are NOT using a SDXL model ckpt right?

Wrong - I'm using SDXL.
On non-sdxl (original) backend it works fine after adding DictWithShape to prompt_parser

the vlad's may come up with it later (?)

Not so. Vlad's fork uses diffusers backend for SDXL - it works differently

and check if it works for you temporarily

Yes, tried it already, but have the same AttributeError: 'NoneType' object has no attribute 'shape' but differently showed.
I'll provide it soon

n_stages: 2
n_frames: 32
steps: [30]
seed: 2919668836
subseed: -1.0
subseed_strength: 0
[stage 1/2]
  pos prompt: a dog
  neg prompt:
Progress  2.35it/s ██████████████████████████████████████ 100% 40/40 00:16 00:00
[stage 2/2]
  pos prompt: a cat
  neg prompt:
Progress  2.55it/s ██████████████████████████████████████ 100% 40/40 00:15 00:00
Traceback (most recent call last):
  File "D:\auto-latest\automatic\extensions\stable-diffusion-webui-prompt-travel\scripts\prompt_travel.py", line 610, in run
    runner()
  File "D:\auto-latest\automatic\extensions\stable-diffusion-webui-prompt-travel\scripts\prompt_travel.py", line 670, in run_linear
    inter_pos_hidden.value = lerp_fn(from_pos_hidden.value, to_pos_hidden.value, alpha)
  File "D:\auto-latest\automatic\extensions\stable-diffusion-webui-prompt-travel\scripts\prompt_travel.py", line 153, in wrapper
    condA, condB = cond_align(condA, condB)
  File "D:\auto-latest\automatic\extensions\stable-diffusion-webui-prompt-travel\scripts\prompt_travel.py", line 149, in cond_align
    condA, condB = align_tensor(condA, condB)
  File "D:\auto-latest\automatic\extensions\stable-diffusion-webui-prompt-travel\scripts\prompt_travel.py", line 140, in align_tensor
    d = x.shape[0] - y.shape[0]
AttributeError: 'NoneType' object has no attribute 'shape'

[sebaxakerhtc]

@Kahsolt As you can see - it generates first image, then a second and stops with this

[sebaxakerhtc]

@Kahsolt
I'm looking at the script and found this
from modules.processing
But there's a new file named processing_diffusers.py which I think used instead of processing.py on diffusers backend with SDXL

[sebaxakerhtc]

@Kahsolt thank you for your time and a great work!

[Kahsolt]

Well, sadly to confirm that this diffusers backend is very simplified, and it does not utilize the original callback system (see stuff in modules.script_callbacks), so we can NOT peep the condition tensor for interpolation (got 'NoneType' error)

And yet good news... you're always welcome back to A1111 lol :)

[sebaxakerhtc]

Well, sadly to confirm that this diffusers backend is very simplified, and it does not utilize the original callback system (see stuff in modules.script_callbacks), so we can NOT peep the condition tensor for interpolation (got 'NoneType' error)

And yet good news... you're always welcome back to A1111 lol :)

Do that this morning only for prompt-travel :)

btw, on original backend it works fine with commit 98a6523