[API] Add multi-vibe and fix default v2/v3 uc preset

example/generate_image_with_vibe.py

@@ -0,0 +1,64 @@
+"""
+{filename}
+==============================================================================
+
+| Example of how to generate an image with inpainting
+|
+| The resulting image will be placed in a folder named "results"
+"""
+
+import asyncio
+import base64
+from pathlib import Path
+
+from example.boilerplate import API
+from novelai_api.ImagePreset import ImageGenerationType, ImageModel, ImagePreset
+
+
+async def single_vibe():
+    d = Path("results")
+    d.mkdir(exist_ok=True)
+
+    async with API() as api_handler:
+        api = api_handler.api
+
+        image = base64.b64encode((d / "image.png").read_bytes()).decode()
+
+        model = ImageModel.Anime_v3
+
+        preset = ImagePreset.from_default_config(model)
+        preset.reference_image = image
+        preset.reference_strength = 0.6
+        preset.reference_information_extracted = 1.0
+        preset.seed = 42
+
+        async for _, img in api.high_level.generate_image("1girl", model, preset, ImageGenerationType.NORMAL):
+            (d / "image_with_vibe.png").write_bytes(img)
+
+
+async def multi_vibe():
+    d = Path("results")
+    d.mkdir(exist_ok=True)
+
+    async with API() as api_handler:
+        api = api_handler.api
+
+        image1 = base64.b64encode((d / "image1.png").read_bytes()).decode()
+        image2 = base64.b64encode((d / "image2.png").read_bytes()).decode()
+        image3 = base64.b64encode((d / "image3.png").read_bytes()).decode()
+
+        model = ImageModel.Anime_v3
+
+        preset = ImagePreset.from_default_config(model)
+        preset.reference_image_multiple = [image1, image2, image3]
+        preset.reference_strength_multiple = [0.6, 0.6, 0.6]
+        preset.reference_information_extracted_multiple = [1.0, 1.0, 1.0]
+        preset.seed = 42
+
+        async for _, img in api.high_level.generate_image("1girl", model, preset, ImageGenerationType.NORMAL):
+            (d / "image_with_multivibe.png").write_bytes(img)
+
+
+if __name__ == "__main__":
+    asyncio.run(single_vibe())
+    asyncio.run(multi_vibe())

novelai_api/ImagePreset.py

@@ -7,7 +7,7 @@
 import sys
 import typing
 from pathlib import Path
-from typing import Any, Dict, Optional, Tuple, Union
+from typing import Any, Dict, List, Optional, Tuple, Union
 
 from novelai_api.ImagePreset_CostTables import DDIM_COSTS, NAI_COSTS, SMEA_COSTS, SMEA_DYN_COSTS
 from novelai_api.python_utils import NoneType, expand_kwargs
@@ -262,6 +262,12 @@ class ImagePreset:
     reference_information_extracted: float
     #: https://docs.novelai.net/.image/vibetransfer.html#reference-strength
     reference_strength: float
+    #: reference_image for multi-vibe transfer
+    reference_image_multiple: List[str]
+    #: reference_information_extracted for multi-vibe transfer
+    reference_information_extracted_multiple: List[float]
+    #: reference_strength for multi-vibe transfer
+    reference_strength_multiple: List[float]
 
     #: Use the old behavior of prompt separation at the 75 tokens mark (can cut words in half)
     legacy_v3_extend: bool

novelai_api/image_presets/presets_v2/default.preset

@@ -6,7 +6,7 @@
     "n_samples": 1,
     "strength": 0.7,
     "noise": 0,
-    "uc_preset": "Preset_Low_Quality_Bad_Anatomy",
+    "uc_preset": "Preset_Heavy",
     "quality_toggle": true,
     "auto_smea": true,
     "smea": false,

novelai_api/image_presets/presets_v3/default.preset

@@ -6,7 +6,7 @@
     "n_samples": 1,
     "strength": 0.7,
     "noise": 0,
-    "uc_preset": "Preset_Low_Quality_Bad_Anatomy",
+    "uc_preset": "Preset_Heavy",
     "quality_toggle": true,
     "auto_smea": true,
     "smea": false,
@@ -19,8 +19,6 @@
     "cfg_rescale": 0,
     "noise_schedule": "native",
     "legacy_v3_extend": false,
-    "reference_information_extracted": 1,
-    "reference_strength": 0.6,
     "params_version": 1,
 
     "seed": 0,

tests/api/test_imagegen_samplers.py

@@ -39,7 +39,7 @@ async def test_samplers(
     model, sampler = model_sampler
 
     # ddim_v3 only work with Anime v3
-    if model not in (ImageModel.Anime_v3,) and sampler is ImageSampler.ddim_v3:
+    if sampler is ImageSampler.ddim_v3 and model not in (ImageModel.Anime_v3,):
         return
 
     logger = api_handle.logger
@@ -52,9 +52,6 @@ async def test_samplers(
     # Furry doesn't have UCPreset.Preset_Low_Quality_Bad_Anatomy
     if model is ImageModel.Furry:
         preset.uc_preset = UCPreset.Preset_Low_Quality
-    # Anime v2 uses different presets than v1
-    elif model in (ImageModel.Anime_v2, ImageModel.Anime_v3):
-        preset.uc_preset = UCPreset.Preset_Heavy
 
     async for _, _ in api.high_level.generate_image("1girl", model, preset):
         pass