pass extra width height
Browse files- app-img2img.py +20 -18
- app-txt2img.py +19 -17
- img2img/index.html +3 -2
- requirements.txt +1 -0
app-img2img.py
CHANGED
|
@@ -21,10 +21,11 @@ import os
|
|
| 21 |
import time
|
| 22 |
import psutil
|
| 23 |
|
| 24 |
-
|
| 25 |
MAX_QUEUE_SIZE = int(os.environ.get("MAX_QUEUE_SIZE", 0))
|
| 26 |
TIMEOUT = float(os.environ.get("TIMEOUT", 0))
|
| 27 |
SAFETY_CHECKER = os.environ.get("SAFETY_CHECKER", None)
|
|
|
|
|
|
|
| 28 |
|
| 29 |
# check if MPS is available OSX only M1/M2/M3 chips
|
| 30 |
mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
|
|
@@ -56,7 +57,7 @@ else:
|
|
| 56 |
custom_revision="main",
|
| 57 |
)
|
| 58 |
pipe.vae = AutoencoderTiny.from_pretrained(
|
| 59 |
-
"madebyollin/taesd", torch_dtype=
|
| 60 |
)
|
| 61 |
pipe.set_progress_bar_config(disable=True)
|
| 62 |
pipe.to(torch_device=torch_device, torch_dtype=torch_dtype).to(device)
|
|
@@ -77,18 +78,29 @@ compel_proc = Compel(
|
|
| 77 |
user_queue_map = {}
|
| 78 |
|
| 79 |
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
|
| 84 |
num_inference_steps = 3
|
| 85 |
results = pipe(
|
| 86 |
prompt_embeds=prompt_embeds,
|
| 87 |
generator=generator,
|
| 88 |
image=input_image,
|
| 89 |
-
strength=strength,
|
| 90 |
num_inference_steps=num_inference_steps,
|
| 91 |
-
guidance_scale=guidance_scale,
|
|
|
|
|
|
|
| 92 |
lcm_origin_steps=50,
|
| 93 |
output_type="pil",
|
| 94 |
)
|
|
@@ -112,13 +124,6 @@ app.add_middleware(
|
|
| 112 |
)
|
| 113 |
|
| 114 |
|
| 115 |
-
class InputParams(BaseModel):
|
| 116 |
-
seed: int
|
| 117 |
-
prompt: str
|
| 118 |
-
strength: float
|
| 119 |
-
guidance_scale: float
|
| 120 |
-
|
| 121 |
-
|
| 122 |
@app.websocket("/ws")
|
| 123 |
async def websocket_endpoint(websocket: WebSocket):
|
| 124 |
await websocket.accept()
|
|
@@ -177,10 +182,7 @@ async def stream(user_id: uuid.UUID):
|
|
| 177 |
|
| 178 |
image = predict(
|
| 179 |
input_image,
|
| 180 |
-
params
|
| 181 |
-
params.guidance_scale,
|
| 182 |
-
params.strength,
|
| 183 |
-
params.seed,
|
| 184 |
)
|
| 185 |
if image is None:
|
| 186 |
continue
|
|
|
|
| 21 |
import time
|
| 22 |
import psutil
|
| 23 |
|
|
|
|
| 24 |
MAX_QUEUE_SIZE = int(os.environ.get("MAX_QUEUE_SIZE", 0))
|
| 25 |
TIMEOUT = float(os.environ.get("TIMEOUT", 0))
|
| 26 |
SAFETY_CHECKER = os.environ.get("SAFETY_CHECKER", None)
|
| 27 |
+
WIDTH = 512
|
| 28 |
+
HEIGHT = 512
|
| 29 |
|
| 30 |
# check if MPS is available OSX only M1/M2/M3 chips
|
| 31 |
mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
|
|
|
|
| 57 |
custom_revision="main",
|
| 58 |
)
|
| 59 |
pipe.vae = AutoencoderTiny.from_pretrained(
|
| 60 |
+
"madebyollin/taesd", torch_dtype=torch_dtype, use_safetensors=True
|
| 61 |
)
|
| 62 |
pipe.set_progress_bar_config(disable=True)
|
| 63 |
pipe.to(torch_device=torch_device, torch_dtype=torch_dtype).to(device)
|
|
|
|
| 78 |
user_queue_map = {}
|
| 79 |
|
| 80 |
|
| 81 |
+
class InputParams(BaseModel):
|
| 82 |
+
prompt: str
|
| 83 |
+
seed: int = 2159232
|
| 84 |
+
guidance_scale: float = 8.0
|
| 85 |
+
strength: float = 0.5
|
| 86 |
+
width: int = WIDTH
|
| 87 |
+
height: int = HEIGHT
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
def predict(input_image: Image.Image, params: InputParams):
|
| 91 |
+
generator = torch.manual_seed(params.seed)
|
| 92 |
+
prompt_embeds = compel_proc(params.prompt)
|
| 93 |
# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
|
| 94 |
num_inference_steps = 3
|
| 95 |
results = pipe(
|
| 96 |
prompt_embeds=prompt_embeds,
|
| 97 |
generator=generator,
|
| 98 |
image=input_image,
|
| 99 |
+
strength=params.strength,
|
| 100 |
num_inference_steps=num_inference_steps,
|
| 101 |
+
guidance_scale=params.guidance_scale,
|
| 102 |
+
width=params.width,
|
| 103 |
+
height=params.height,
|
| 104 |
lcm_origin_steps=50,
|
| 105 |
output_type="pil",
|
| 106 |
)
|
|
|
|
| 124 |
)
|
| 125 |
|
| 126 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
@app.websocket("/ws")
|
| 128 |
async def websocket_endpoint(websocket: WebSocket):
|
| 129 |
await websocket.accept()
|
|
|
|
| 182 |
|
| 183 |
image = predict(
|
| 184 |
input_image,
|
| 185 |
+
params,
|
|
|
|
|
|
|
|
|
|
| 186 |
)
|
| 187 |
if image is None:
|
| 188 |
continue
|
app-txt2img.py
CHANGED
|
@@ -25,7 +25,8 @@ import psutil
|
|
| 25 |
MAX_QUEUE_SIZE = int(os.environ.get("MAX_QUEUE_SIZE", 0))
|
| 26 |
TIMEOUT = float(os.environ.get("TIMEOUT", 0))
|
| 27 |
SAFETY_CHECKER = os.environ.get("SAFETY_CHECKER", None)
|
| 28 |
-
|
|
|
|
| 29 |
# check if MPS is available OSX only M1/M2/M3 chips
|
| 30 |
mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
|
| 31 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
@@ -66,9 +67,9 @@ pipe.unet.to(memory_format=torch.channels_last)
|
|
| 66 |
if psutil.virtual_memory().total < 64 * 1024**3:
|
| 67 |
pipe.enable_attention_slicing()
|
| 68 |
|
| 69 |
-
if not mps_available:
|
| 70 |
-
|
| 71 |
-
|
| 72 |
|
| 73 |
compel_proc = Compel(
|
| 74 |
tokenizer=pipe.tokenizer,
|
|
@@ -77,17 +78,25 @@ compel_proc = Compel(
|
|
| 77 |
)
|
| 78 |
user_queue_map = {}
|
| 79 |
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
|
| 85 |
num_inference_steps = 8
|
| 86 |
results = pipe(
|
| 87 |
prompt_embeds=prompt_embeds,
|
| 88 |
generator=generator,
|
| 89 |
num_inference_steps=num_inference_steps,
|
| 90 |
-
guidance_scale=guidance_scale,
|
|
|
|
|
|
|
| 91 |
lcm_origin_steps=50,
|
| 92 |
output_type="pil",
|
| 93 |
)
|
|
@@ -110,13 +119,6 @@ app.add_middleware(
|
|
| 110 |
allow_headers=["*"],
|
| 111 |
)
|
| 112 |
|
| 113 |
-
|
| 114 |
-
class InputParams(BaseModel):
|
| 115 |
-
prompt: str
|
| 116 |
-
seed: int
|
| 117 |
-
guidance_scale: float
|
| 118 |
-
|
| 119 |
-
|
| 120 |
@app.websocket("/ws")
|
| 121 |
async def websocket_endpoint(websocket: WebSocket):
|
| 122 |
await websocket.accept()
|
|
@@ -173,7 +175,7 @@ async def stream(user_id: uuid.UUID):
|
|
| 173 |
if params is None:
|
| 174 |
continue
|
| 175 |
|
| 176 |
-
image = predict(params
|
| 177 |
if image is None:
|
| 178 |
continue
|
| 179 |
frame_data = io.BytesIO()
|
|
|
|
| 25 |
MAX_QUEUE_SIZE = int(os.environ.get("MAX_QUEUE_SIZE", 0))
|
| 26 |
TIMEOUT = float(os.environ.get("TIMEOUT", 0))
|
| 27 |
SAFETY_CHECKER = os.environ.get("SAFETY_CHECKER", None)
|
| 28 |
+
WIDTH = 512
|
| 29 |
+
HEIGHT = 512
|
| 30 |
# check if MPS is available OSX only M1/M2/M3 chips
|
| 31 |
mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
|
| 32 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
|
| 67 |
if psutil.virtual_memory().total < 64 * 1024**3:
|
| 68 |
pipe.enable_attention_slicing()
|
| 69 |
|
| 70 |
+
# if not mps_available:
|
| 71 |
+
# pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
|
| 72 |
+
# pipe(prompt="warmup", num_inference_steps=1, guidance_scale=8.0)
|
| 73 |
|
| 74 |
compel_proc = Compel(
|
| 75 |
tokenizer=pipe.tokenizer,
|
|
|
|
| 78 |
)
|
| 79 |
user_queue_map = {}
|
| 80 |
|
| 81 |
+
class InputParams(BaseModel):
|
| 82 |
+
prompt: str
|
| 83 |
+
seed: int = 2159232
|
| 84 |
+
guidance_scale: float = 8.0
|
| 85 |
+
width: int = WIDTH
|
| 86 |
+
height: int = HEIGHT
|
| 87 |
+
|
| 88 |
+
def predict(params: InputParams):
|
| 89 |
+
generator = torch.manual_seed(params.seed)
|
| 90 |
+
prompt_embeds = compel_proc(params.prompt)
|
| 91 |
# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
|
| 92 |
num_inference_steps = 8
|
| 93 |
results = pipe(
|
| 94 |
prompt_embeds=prompt_embeds,
|
| 95 |
generator=generator,
|
| 96 |
num_inference_steps=num_inference_steps,
|
| 97 |
+
guidance_scale=params.guidance_scale,
|
| 98 |
+
width=params.width,
|
| 99 |
+
height=params.height,
|
| 100 |
lcm_origin_steps=50,
|
| 101 |
output_type="pil",
|
| 102 |
)
|
|
|
|
| 119 |
allow_headers=["*"],
|
| 120 |
)
|
| 121 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 122 |
@app.websocket("/ws")
|
| 123 |
async def websocket_endpoint(websocket: WebSocket):
|
| 124 |
await websocket.accept()
|
|
|
|
| 175 |
if params is None:
|
| 176 |
continue
|
| 177 |
|
| 178 |
+
image = predict(params)
|
| 179 |
if image is None:
|
| 180 |
continue
|
| 181 |
frame_data = io.BytesIO()
|
img2img/index.html
CHANGED
|
@@ -10,8 +10,9 @@
|
|
| 10 |
<script src="https://cdn.jsdelivr.net/npm/[email protected]/piexif.min.js"></script>
|
| 11 |
<script src="https://cdn.tailwindcss.com"></script>
|
| 12 |
<script type="module">
|
| 13 |
-
|
| 14 |
-
const
|
|
|
|
| 15 |
const seedEl = document.querySelector("#seed");
|
| 16 |
const promptEl = document.querySelector("#prompt");
|
| 17 |
const guidanceEl = document.querySelector("#guidance-scale");
|
|
|
|
| 10 |
<script src="https://cdn.jsdelivr.net/npm/[email protected]/piexif.min.js"></script>
|
| 11 |
<script src="https://cdn.tailwindcss.com"></script>
|
| 12 |
<script type="module">
|
| 13 |
+
// you can change the size of the input image to 768x768 if you have a powerful GPU
|
| 14 |
+
const WIDTH = 512;
|
| 15 |
+
const HEIGHT = 512;
|
| 16 |
const seedEl = document.querySelector("#seed");
|
| 17 |
const promptEl = document.querySelector("#prompt");
|
| 18 |
const guidanceEl = document.querySelector("#guidance-scale");
|
requirements.txt
CHANGED
|
@@ -1,6 +1,7 @@
|
|
| 1 |
diffusers==0.21.4
|
| 2 |
transformers==4.34.1
|
| 3 |
gradio==3.50.2
|
|
|
|
| 4 |
torch==2.1.0
|
| 5 |
fastapi==0.104.0
|
| 6 |
uvicorn==0.23.2
|
|
|
|
| 1 |
diffusers==0.21.4
|
| 2 |
transformers==4.34.1
|
| 3 |
gradio==3.50.2
|
| 4 |
+
--extra-index-url https://download.pytorch.org/whl/cu121
|
| 5 |
torch==2.1.0
|
| 6 |
fastapi==0.104.0
|
| 7 |
uvicorn==0.23.2
|