Update README.md
Browse files
README.md
CHANGED
|
@@ -19,7 +19,7 @@ inference, its job is to generate the image latents given text. These image late
|
|
| 19 |
### Image Sizes
|
| 20 |
Würstchen was trained on image resolutions between 1024x1024 & 1536x1536. We sometimes also observe good outputs at resolutions like 1024x2048. Feel free to try it out.
|
| 21 |
We also observed that the Prior (Stage C) adapts extremely fast to new resolutions. So finetuning it at 2048x2048 should be computationally cheap.
|
| 22 |
-
<img src="https://cdn-uploads.huggingface.co/production/uploads/634cb5eefb80cc6bcaf63c3e/
|
| 23 |
|
| 24 |
## How to run
|
| 25 |
This pipeline should be run together with https://huggingface.co/warp-ai/wuerstchen:
|
|
@@ -62,6 +62,12 @@ decoder_output = decoder_pipeline(
|
|
| 62 |
).images
|
| 63 |
```
|
| 64 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
## Model Details
|
| 66 |
- **Developed by:** Pablo Pernias, Dominic Rampas
|
| 67 |
- **Model type:** Diffusion-based text-to-image generation model
|
|
|
|
| 19 |
### Image Sizes
|
| 20 |
Würstchen was trained on image resolutions between 1024x1024 & 1536x1536. We sometimes also observe good outputs at resolutions like 1024x2048. Feel free to try it out.
|
| 21 |
We also observed that the Prior (Stage C) adapts extremely fast to new resolutions. So finetuning it at 2048x2048 should be computationally cheap.
|
| 22 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/634cb5eefb80cc6bcaf63c3e/5pA5KUfGmvsObqiIjdGY1.jpeg" width=1000>
|
| 23 |
|
| 24 |
## How to run
|
| 25 |
This pipeline should be run together with https://huggingface.co/warp-ai/wuerstchen:
|
|
|
|
| 62 |
).images
|
| 63 |
```
|
| 64 |
|
| 65 |
+
### Image Sampling Times
|
| 66 |
+
The figure shows the inference times (on an A100) for different batch sizes (`num_images_per_prompt`) on Würstchen compared to [Stable Diffusion XL](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) (without refiner).
|
| 67 |
+
The left figure shows inference times (using torch > 2.0), whereas the right figure applies `torch.compile` to both pipelines in advance.
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
|
| 71 |
## Model Details
|
| 72 |
- **Developed by:** Pablo Pernias, Dominic Rampas
|
| 73 |
- **Model type:** Diffusion-based text-to-image generation model
|