Please download LDSR to activate more upscale features...

[sedetweiler]

The instructions say... "Please download LDSR to activate more upscale features, instructions are available at the Github" But nowhere on the site is there any sort of a hint on what to do next.

[AmoArt]

+1 to this, I'm going to assume that author simply forgot to update the instructions when updating the code.
Additional, I would also love to know were to get the 'GoLatent' model since that is not included either too.

[devilismyfriend]

+1 to this, I'm going to assume that author simply forgot to update the instructions when updating the code. Additional, I would also love to know were to get the 'GoLatent' model since that is not included either too.

GoLatent is just using GoBig before LDSR, it's not a separate model.

[JoshuaKimsey]

Yeah, I was just fixing to open a discussion for this. @hlky and @devilismyfriend, we're gonna need some instructions on how to activate the LDSR part of the new upscalers addition, like we have for GFPGAN and realESRGAN. I know in the PR it mentioned needing to clone a repo and download a model, but not sure if that held through to the final merge.

Also wanna know how this works so I can update my Linux script to handle this automatically for a user. :)

[darkstorm2150]

Need github Instruction to install Latent Diffusion Super Resolution :D

[mchaker]

It looks like this is the way to get LDSR:

https://github.com/devilismyfriend/latent-diffusion#get-the-models-1

[devilismyfriend]

The instructions say... "Please download LDSR to activate more upscale features, instructions are available at the Github" But nowhere on the site is there any sort of a hint on what to do next.

Yes lol it's a placeholder, sorry, as you can tell this is in active development still.

How to install LDSR:

git clone https://github.com/devilismyfriend/latent-diffusion to your src folder, make sure the name of the folder is "latent-diffusion" and not "latent-diffusion-main".

Go into the latent-diffusion folder and you can use the download_model.bat to download and place the models or you can do it yourself by placing the model.ckpt and project.yaml files under /src/latent-diffusion/experiments/pretrained_models/

[hlky]

can just add this this to environment.yaml
will do it soon, testing the new streamlit version

[devilismyfriend]

cool thanks, I changed the readme and wrote down to git clone the repo, but once you add it to the environment we'll just remove that part.

[churchofthought]

thanks guys, yall deserve some 👏

[sedetweiler]

thanks guys, yall deserve some 👏

I actually gave some cash. Thanks again, guys!

[AmoArt]

the devilismyfriend instructions work, however when running the upscale process 'Latent Diffusion SR' it ignores/overrides my sampling method and steps and sets them up as DDIM 100 steps each run.

[TristanMorelle]

@devilismyfriend
with single and multiple images i get this error
nothing was produced

  Processing images...
Unloaded GFPGAN
Loaded RealESGAN
GoBig upscaling will process a total of 9 images tiled as 3x3 in a total of 9 batches.
[MemMon] Recording max memory usage...

Iteration: 1/1
100%|██████████████████████████████████████████████████████████████████████████████████| 16/16 [00:01<00:00,  8.08it/s]
Traceback (most recent call last):
  File "C:\Users\Tristan\.conda\envs\ldo\lib\site-packages\gradio\routes.py", line 247, in run_predict
    output = await app.blocks.process_api(
  File "..\.conda\envs\ldo\lib\site-packages\gradio\blocks.py", line 641, in process_api
    predictions, duration = await self.call_function(fn_index, processed_input)
  File "..\.conda\envs\ldo\lib\site-packages\gradio\blocks.py", line 556, in call_function
    prediction = await anyio.to_thread.run_sync(
  File "..\.conda\envs\ldo\lib\site-packages\anyio\to_thread.py", line 31, in run_sync
    return await get_asynclib().run_sync_in_worker_thread(
  File ..\.conda\envs\ldo\lib\site-packages\anyio\_backends\_asyncio.py", line 937, in run_sync_in_worker_thread
    return await future
  File ..\.conda\envs\ldo\lib\site-packages\anyio\_backends\_asyncio.py", line 867, in run
    result = context.run(func, *args)
  File "scripts/webui.py", line 1731, in imgproc
    image = processGoBig(image)
  File "scripts/webui.py", line 1665, in processGoBig
    work_results.append(output_images[0])
IndexError: list index out of range
[MemMon] Stopped recording.

[bilalqureshi1992]

even after the LDSR part it's giving me the following error, (any solution??):
RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 3.42 GiB already allocated; 0 bytes free; 3.48 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[mchaker]

Your graphics card is not powerful enough (not enough VRAM).

Have you tried the Doggettx memory optimization? : https://gist.github.com/mchaker/e0cf180db6849fa03696293b3fc8f485

[StantheBrain]

-----RUNNING ON 4GB----- https://rentry.org/GUItard
1- Edit /scripts/relauncher.py
2- Edit line 14
(# Run Stable Diffusion in Optimized Mode - Only requires 4Gb of VRAM, but is significantly slower)
Line 14 change : False to True
3- Save

(this tells it to optimize Vram by generating incrementally, it should be noticed that this sacrifices generation speed)

[bilalqureshi1992]

thanks man that worked like a charm :)

[devilismyfriend]

LDSR requires at least 8GB of ram.

…

On Tue., Sep. 6, 2022, 5:23 p.m. Bilal Qureshi, ***@***.***> wrote: even after the LDSR part it's giving me the following error, (any solution??): RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 3.42 GiB already allocated; 0 bytes free; 3.48 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF — Reply to this email directly, view it on GitHub <#450 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUYC4IE36KU4KR7A4TA6BJTV45HW3ANCNFSM6AAAAAAQCR3HRU> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[bilalqureshi1992]

ok if I don't want to use it, how can I comment it out?

[devilismyfriend]

Just don't use it? It's hardly a default option

…

On Tue., Sep. 6, 2022, 5:28 p.m. Bilal Qureshi, ***@***.***> wrote: ok if I don't want to use it, how can I comment it out? — Reply to this email directly, view it on GitHub <#450 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUYC4IBLO37AGANXCPOARIDV45IHDANCNFSM6AAAAAAQCR3HRU> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>