How to use base models other than SDXL?

[vinjn]

How to use base models other than SDXL?

[sampadsams]

👍 Solution to use other models on Fooocus :

Don't read this :

Models have different types that should be loaded correctly in their special place , a model generates images and another changes or refines it , that's the standard. Some models come with the whole sequence , but in separated files , because they have to be loaded in that standard. Example of that is the default XL model that Fooocus uses , it is one model but comes in two files , and it causes problems if you intend to replace only one of them. Solution, is to replace both.

Read this :

1 . Deleting or disabling these lines in refresh_base_model() function in default_pipeline.py:

if not isinstance(xl_base.unet.model, SDXL): print('Model not supported. Fooocus only support SDXL model as the base model.') xl_base = None xl_base_hash = '' refresh_base_model(modules.path.default_base_model_name) xl_base_hash = name xl_base_patched = xl_base xl_base_patched_hash = '' return

2 . Put a CHECKPOINT TRAINED type model like DreamSharper in checkpoints folder and set it in advanced section (refinder to None)

3. Optional: You can put Lora type model in loras folder and use with your base model

I'll update this as I explore different types of models.

[ujl123]

Advanced checkbox, then advanced tab on the right

[vinjn]

Nope it didn't work and I saw the following error:

Model not supported. Fooocus only support SDXL model as the base model.

[Tigwin]

It would be really nice if we could use a different base for the initial render and then use sdxl refiner or base+refiner to finish it off.

[ujl123]

Nope it didn't work and I saw the following error:

Model not supported. Fooocus only support SDXL model as the base model.

Sorry for misleading you. I had misread your title and thought you wanted other base models of SDXL that wasn't the basic one that SAI released. AFAIK only the base SDXL models work.

[mashb1t]

It would be really nice if we could use a different base for the initial render and then use sdxl refiner or base+refiner to finish it off.

Well, it's been a long time folks, but the answer is exactly this: #1568 (comment)

[Riohigh6387]

👍 Solution to use other models on Fooocus :

Don't read this :

Models have different types that should be loaded correctly in their special place , a model generates images and another changes or refines it , that's the standard. Some models come with the whole sequence , but in separated files , because they have to be loaded in that standard. Example of that is the default XL model that Fooocus uses , it is one model but comes in two files , and it causes problems if you intend to replace only one of them. Solution, is to replace both.

Read this :

1 . Deleting or disabling these lines in refresh_base_model() function in default_pipeline.py:

if not isinstance(xl_base.unet.model, SDXL): print('Model not supported. Fooocus only support SDXL model as the base model.') xl_base = None xl_base_hash = '' refresh_base_model(modules.path.default_base_model_name) xl_base_hash = name xl_base_patched = xl_base xl_base_patched_hash = '' return

2 . Put a CHECKPOINT TRAINED type model like DreamSharper in checkpoints folder and set it in advanced section (refinder to None)

3. Optional: You can put Lora type model in loras folder and use with your base model

I'll update this as I explore different types of models.

Any updates? i did exactly as you said and it worked fine till it didnt , wampwamp.black screen at finish.. Do you know what i can do to prevent this? anything helps, i like to tinker. Thanks in advance.