Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
import gradio as gr
|
2 |
import jax
|
|
|
3 |
from flax.jax_utils import replicate
|
4 |
from flax.training.common_utils import shard
|
5 |
-
from diffusers import FlaxStableDiffusionPipeline
|
6 |
|
7 |
pipeline, pipeline_params = FlaxStableDiffusionPipeline.from_pretrained(
|
8 |
"bguisard/stable-diffusion-nano",
|
@@ -13,11 +13,11 @@ def generate_image(prompt: str, inference_steps: int = 30, prng_seed: int = 0):
|
|
13 |
rng = jax.random.PRNGKey(int(prng_seed))
|
14 |
rng = jax.random.split(rng, jax.device_count())
|
15 |
p_params = replicate(pipeline_params)
|
16 |
-
|
17 |
num_samples = 1
|
18 |
prompt_ids = pipeline.prepare_inputs([prompt] * num_samples)
|
19 |
prompt_ids = shard(prompt_ids)
|
20 |
-
|
21 |
images = pipeline(
|
22 |
prompt_ids=prompt_ids,
|
23 |
params=p_params,
|
@@ -30,7 +30,7 @@ def generate_image(prompt: str, inference_steps: int = 30, prng_seed: int = 0):
|
|
30 |
|
31 |
images = images.reshape((num_samples,) + images.shape[-3:])
|
32 |
images = pipeline.numpy_to_pil(images)
|
33 |
-
return images
|
34 |
|
35 |
|
36 |
prompt_input = gr.inputs.Textbox(
|
@@ -44,7 +44,7 @@ seed_input = gr.inputs.Number(default=0, label="Seed")
|
|
44 |
app = gr.Interface(
|
45 |
fn=generate_image,
|
46 |
inputs=[prompt_input, inf_steps_input, seed_input],
|
47 |
-
outputs=
|
48 |
title="Stable Diffusion Nano",
|
49 |
description=(
|
50 |
"Based on stable diffusion and fine-tuned on 128x128 images, "
|
|
|
1 |
import gradio as gr
|
2 |
import jax
|
3 |
+
from diffusers import FlaxStableDiffusionPipeline
|
4 |
from flax.jax_utils import replicate
|
5 |
from flax.training.common_utils import shard
|
|
|
6 |
|
7 |
pipeline, pipeline_params = FlaxStableDiffusionPipeline.from_pretrained(
|
8 |
"bguisard/stable-diffusion-nano",
|
|
|
13 |
rng = jax.random.PRNGKey(int(prng_seed))
|
14 |
rng = jax.random.split(rng, jax.device_count())
|
15 |
p_params = replicate(pipeline_params)
|
16 |
+
|
17 |
num_samples = 1
|
18 |
prompt_ids = pipeline.prepare_inputs([prompt] * num_samples)
|
19 |
prompt_ids = shard(prompt_ids)
|
20 |
+
|
21 |
images = pipeline(
|
22 |
prompt_ids=prompt_ids,
|
23 |
params=p_params,
|
|
|
30 |
|
31 |
images = images.reshape((num_samples,) + images.shape[-3:])
|
32 |
images = pipeline.numpy_to_pil(images)
|
33 |
+
return images[0]
|
34 |
|
35 |
|
36 |
prompt_input = gr.inputs.Textbox(
|
|
|
44 |
app = gr.Interface(
|
45 |
fn=generate_image,
|
46 |
inputs=[prompt_input, inf_steps_input, seed_input],
|
47 |
+
outputs="image",
|
48 |
title="Stable Diffusion Nano",
|
49 |
description=(
|
50 |
"Based on stable diffusion and fine-tuned on 128x128 images, "
|