feat: add more models and sideload

app.py

@@ -25,13 +25,22 @@ import modules.safe as _
 from modules.lora import LoRANetwork
 
 models = [
-    ("AbyssOrangeMix2", "Korakoe/AbyssOrangeMix2-HF"),
-    ("Basil Mix", "nuigurumi/basil_mix"),
-    ("Pastal Mix", "andite/pastel-mix"),
+    ("AbyssOrangeMix2 (fast)", "Korakoe/AbyssOrangeMix2-HF", 2),
+    ("Pastal Mix (fast)", "andite/pastel-mix", 2),
 ]
 
-base_name, base_model,  = models[0]
-clip_skip = 2
+alt_models = [
+    ("ACertainModel", "JosephusCheung/ACertainModel", 2),
+    ("Basil Mix", "nuigurumi/basil_mix", 2),
+    ("Stable Diffusion V1.5", "runwayml/stable-diffusion-v1-5", 1),
+    ("Anything V3.0", "Linaqruf/anything-v3.0", 2),
+    ("Open Journey", "prompthero/openjourney", 1),
+    ("Eimis AnimeDiffusion", "eimiss/EimisAnimeDiffusion_1.0v", 2)
+    ("Dreamlike Photoreal 2.0", "dreamlike-art/dreamlike-photoreal-2.0", 1)
+    ("Redshift Diffusion", "nitrosocke/redshift-diffusion", 1)
+]
+
+base_name, base_model, clip_skip = models[0]
 
 samplers_k_diffusion = [
     ("Euler a", "sample_euler_ancestral", {}),
@@ -91,11 +100,9 @@ pipe = StableDiffusionPipeline(
 )
 
 unet.set_attn_processor(CrossAttnProcessor)
-pipe.set_clip_skip(clip_skip)
 if torch.cuda.is_available():
     pipe = pipe.to("cuda")
 
-
 def get_model_list():
     return models
 
@@ -128,10 +135,10 @@ def get_model(name):
     g_lora = lora_cache[name]
     g_unet.set_attn_processor(CrossAttnProcessor())
     g_lora.reset()
-    return g_unet, g_lora
+    return g_unet, g_lora, models[keys.index(name)][2]
 
 # precache on huggingface
-for model in get_model_list():
+for model in models:
     get_model(model[0])
 
 def error_str(error, title="Error"):
@@ -214,7 +221,8 @@ def inference(
     restore_all()
     generator = torch.Generator("cuda").manual_seed(int(seed))
 
-    local_unet, local_lora = get_model(model)
+    local_unet, local_lora, clip_skip = get_model(model)
+    pipe.set_clip_skip(clip_skip)
     if lora_state is not None and lora_state != "":
         local_lora.load(lora_state, lora_scale)
         local_lora.to(local_unet.device, dtype=local_unet.dtype)
@@ -279,6 +287,9 @@ def inference(
     if embs is not None and len(embs) > 0:
         restore_all()
 
+    if model in [key[0] for key in alt_models]:
+        local_unet.to("cpu")
+
     end_time = time.time()
     vram_free, vram_total = torch.cuda.mem_get_info()
     print(f"done: res={width}x{height}, step={steps}, time={round(end_time-start_time, 2)}s, vram_alloc={convert_size(vram_total-vram_free)}/{convert_size(vram_total)}")