multiple models loaded at once

[4lt3r3go]

it would be a great thing to be able to generate an image and then send to a sort of img2img that process the same image with another model, already loaded (for who have enough Vram this may be duable)
i'm doing this everytime, with 2 different install of stable diffusions and copy back and forth images.
Imagin doing this directly in the WebUI
Best images i've got so far were all processed with multiple models, at different denoises

[4lt3r3go]

and before you answer: NO: switching between models is not convenient for multiple resons:

1. loading times (not everyone has super fast nvme
2. finding the model is always a pain in the ass if you have a list of infinite models
   Doing this for every image is a no no

[BunnyViking]

if you have enough VRAM i think you can cache models which would be faster for swapping. Which would also suggest what you're suggesting should be possible.

[Kilvoctu]

Set this to a value greater than 0, for each checkpoint you want cached. It will let you swap between that many recent checkpoints instantly. It uses your system RAM (not VRAM), so if you have a good amount of RAM you can cache quite a few checkpoints.

[NasBeru]

how can I load all model in RAM as the server starts? Is there any config param to set for this

[4lt3r3go]

i just discovered i can just open a second instance of SD normally and will create another ip to use it in a different tab

[zx96-001]

how would it be possible to have such a situation?

Size: 512x768, Seed: 123456789, Model: 0.5(nameofthemodel1) + 0.5(nameofthemodel2), Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 12, Model hash: adb7a198fe, Hires steps: 10, Hires upscale: 2, Hires upscaler: Latent (nearest-exact), Denoising strength: 0.55

is it possible to have both model1 and model2 loaded in the vram and SD renders the image with both models loaded with weights of 50% each? i can't understand how an image I saw was generated and I pasted here the relevant part, removing the seed/names of the models just to understand the process but it is a prompt I copied of an image that exists and that apparently was generated using 2 models.

any help/info would be appreciated

[Kadah]

That would be model merging. An extension like SuperMerge would allow you to mix 2-3 models in VRAM without the need to save save the mixed model to disk.