-
Notifications
You must be signed in to change notification settings - Fork 5.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add preset for SDXL Turbo (model DreamShaperXL_Turbo) #1580
base: main
Are you sure you want to change the base?
Conversation
includes default_override_step = 6
I think this preset is not as good as a performance selection, 'cause it overrides the number of steps |
add here Fooocus/modules/async_worker.py Line 173 in 7b5bced
if performance_selection == 'Turbo Speed':
print('Enter SDXL Turbo mode.')
progressbar(async_task, 1, 'Downloading SDXL Turbo components ...')
# change this to config.py could be better
modules.config.load_file_from_url(
url='https://huggingface.co/Lykon/dreamshaper-xl-turbo/resolve/main/DreamShaperXL_Turbo_dpmppSdeKarras_half_pruned_6.safetensors',
model_dir=modules.config.path_checkpoints,
file_name='DreamShaperXL_Turbo_dpmppSdeKarras_half_pruned_6.safetensors'
)
refiner_model_name = 'None'
sampler_name = advanced_parameters.sampler_name = 'dpmpp_sde'
cfg_scale = guidance_scale = 2.0
modules.patch.sharpness = sharpness = 3.0
steps = 6 |
I think presets are better than performance selections, because LCM performance choices have the potential to lock parameters. |
this is also something which i struggled with in the beginning, but i decided to go for a preset as performance selections are not bound to a specific model but rather universally applicable while presets can be very specific and easier extended/changed by users without having to fiddle with the code. |
hardcoding Turbo into project codebase is still under license investigation since some online commercial service also use fooocus for common users wheras Turbo is kind of NC model |
Any update on this? A turbo preset would make it so much easier to work with the new dreamshaper model. |
@RahulVivekNair as you can see there isnt yet an update on the matter. Feel free to use the preset locally or use my fork https://github.com/mashb1t/Fooocus, where the feature has already been merged. |
Switched to your fork, works quite well thanks! |
closes #1499
discussion in #1060
references
examples: https://wiki.mindrenders.com/books/fooocus/chapter/fooocus-image-sets
What's wondering me is that the list "possible_preset_keys" (config.py) isn't used when loading presets, see https://github.com/lllyasviel/Fooocus/blob/main/modules/config.py#L82-L95,
config_dict.update(json.load(json_file))
.In this case the behaviour good, so one can simply add default_override_step to the preset and it's working perfectly fine.
Please let me know if you'd rather also have the key default_override_step in possible_preset_keys.
I'd also propose to add a license hint to the readme if/when adding turbo to the list of presets and to the installer.
Example from https://civitai.com/models/215418:
Thanks!