Use bfloat16, 1 image, no grid (#1)
Browse files- Use bfloat16, 1 image, no grid (75ba64df6af4017799c72e44f57a7e60ac010e68)
Co-authored-by: Pedro Cuenca <[email protected]>
app.py
CHANGED
|
@@ -7,48 +7,30 @@ from flax.training.common_utils import shard
|
|
| 7 |
from PIL import Image
|
| 8 |
from diffusers import FlaxStableDiffusionControlNetPipeline, FlaxControlNetModel
|
| 9 |
import cv2
|
| 10 |
-
import os
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
def image_grid(imgs, rows, cols):
|
| 16 |
-
w, h = imgs[0].size
|
| 17 |
-
grid = Image.new("RGB", size=(cols * w, rows * h))
|
| 18 |
-
for i, img in enumerate(imgs):
|
| 19 |
-
grid.paste(img, box=(i % cols * w, i // cols * h))
|
| 20 |
-
return grid
|
| 21 |
-
|
| 22 |
|
| 23 |
def create_key(seed=0):
|
| 24 |
return jax.random.PRNGKey(seed)
|
| 25 |
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
def canny_filter(image):
|
| 30 |
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
| 31 |
-
|
| 32 |
blurred_image = cv2.GaussianBlur(gray_image, (5, 5), 0)
|
| 33 |
edges_image = cv2.Canny(blurred_image, 50, 150)
|
| 34 |
return edges_image
|
| 35 |
|
| 36 |
# load control net and stable diffusion v1-5
|
| 37 |
controlnet, controlnet_params = FlaxControlNetModel.from_pretrained(
|
| 38 |
-
"jax-diffusers-event/canny-coyo1m", dtype=jnp.
|
| 39 |
)
|
| 40 |
pipe, params = FlaxStableDiffusionControlNetPipeline.from_pretrained(
|
| 41 |
-
"runwayml/stable-diffusion-v1-5",
|
| 42 |
)
|
| 43 |
|
| 44 |
def infer(prompts, negative_prompts, image):
|
| 45 |
-
|
| 46 |
-
|
| 47 |
params["controlnet"] = controlnet_params
|
| 48 |
|
| 49 |
-
num_samples = jax.device_count(
|
| 50 |
rng = create_key(0)
|
| 51 |
-
rng = jax.random.split(rng, jax.device_count(
|
| 52 |
im = canny_filter(image)
|
| 53 |
canny_image = Image.fromarray(im)
|
| 54 |
|
|
@@ -56,12 +38,15 @@ def infer(prompts, negative_prompts, image):
|
|
| 56 |
negative_prompt_ids = pipe.prepare_text_inputs([negative_prompts] * num_samples)
|
| 57 |
processed_image = pipe.prepare_image_inputs([canny_image] * num_samples)
|
| 58 |
|
| 59 |
-
|
|
|
|
|
|
|
|
|
|
| 60 |
|
| 61 |
output = pipe(
|
| 62 |
prompt_ids=prompt_ids,
|
| 63 |
image=processed_image,
|
| 64 |
-
params=
|
| 65 |
prng_seed=rng,
|
| 66 |
num_inference_steps=50,
|
| 67 |
neg_prompt_ids=negative_prompt_ids,
|
|
@@ -69,7 +54,6 @@ def infer(prompts, negative_prompts, image):
|
|
| 69 |
).images
|
| 70 |
|
| 71 |
output_images = pipe.numpy_to_pil(np.asarray(output.reshape((num_samples,) + output.shape[-3:])))
|
| 72 |
-
output_images = image_grid(output_images, num_samples // 4, 4)
|
| 73 |
return output_images
|
| 74 |
|
| 75 |
gr.Interface(infer, inputs=["text", "text", "image"], outputs="gallery").launch()
|
|
|
|
| 7 |
from PIL import Image
|
| 8 |
from diffusers import FlaxStableDiffusionControlNetPipeline, FlaxControlNetModel
|
| 9 |
import cv2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
def create_key(seed=0):
|
| 12 |
return jax.random.PRNGKey(seed)
|
| 13 |
|
|
|
|
|
|
|
|
|
|
| 14 |
def canny_filter(image):
|
| 15 |
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
|
|
|
| 16 |
blurred_image = cv2.GaussianBlur(gray_image, (5, 5), 0)
|
| 17 |
edges_image = cv2.Canny(blurred_image, 50, 150)
|
| 18 |
return edges_image
|
| 19 |
|
| 20 |
# load control net and stable diffusion v1-5
|
| 21 |
controlnet, controlnet_params = FlaxControlNetModel.from_pretrained(
|
| 22 |
+
"jax-diffusers-event/canny-coyo1m", dtype=jnp.bfloat16
|
| 23 |
)
|
| 24 |
pipe, params = FlaxStableDiffusionControlNetPipeline.from_pretrained(
|
| 25 |
+
"runwayml/stable-diffusion-v1-5", controlnet=controlnet, revision="flax", dtype=jnp.bfloat16
|
| 26 |
)
|
| 27 |
|
| 28 |
def infer(prompts, negative_prompts, image):
|
|
|
|
|
|
|
| 29 |
params["controlnet"] = controlnet_params
|
| 30 |
|
| 31 |
+
num_samples = 1 #jax.device_count()
|
| 32 |
rng = create_key(0)
|
| 33 |
+
rng = jax.random.split(rng, jax.device_count())
|
| 34 |
im = canny_filter(image)
|
| 35 |
canny_image = Image.fromarray(im)
|
| 36 |
|
|
|
|
| 38 |
negative_prompt_ids = pipe.prepare_text_inputs([negative_prompts] * num_samples)
|
| 39 |
processed_image = pipe.prepare_image_inputs([canny_image] * num_samples)
|
| 40 |
|
| 41 |
+
p_params = replicate(params)
|
| 42 |
+
prompt_ids = shard(prompt_ids)
|
| 43 |
+
negative_prompt_ids = shard(negative_prompt_ids)
|
| 44 |
+
processed_image = shard(processed_image)
|
| 45 |
|
| 46 |
output = pipe(
|
| 47 |
prompt_ids=prompt_ids,
|
| 48 |
image=processed_image,
|
| 49 |
+
params=p_params,
|
| 50 |
prng_seed=rng,
|
| 51 |
num_inference_steps=50,
|
| 52 |
neg_prompt_ids=negative_prompt_ids,
|
|
|
|
| 54 |
).images
|
| 55 |
|
| 56 |
output_images = pipe.numpy_to_pil(np.asarray(output.reshape((num_samples,) + output.shape[-3:])))
|
|
|
|
| 57 |
return output_images
|
| 58 |
|
| 59 |
gr.Interface(infer, inputs=["text", "text", "image"], outputs="gallery").launch()
|