Commit ·
3f9b2c3
1
Parent(s): 7abb7e9
link to benchmark
Browse files
README.md
CHANGED
|
@@ -33,11 +33,13 @@ Under the hood, it is a lightweight **Rectified Flow** model with **59M** parame
|
|
| 33 |
|
| 34 |
Features:
|
| 35 |
|
| 36 |
-
* Upscaling
|
| 37 |
* The model is trained for **2X** upscaling, but multiple passes can be chained to reach up to **8K** resolution
|
| 38 |
* A full pipeline with Flux generation, upscaling to **8K**, and decoding runs in just **25 seconds** (on RTX 5090)
|
| 39 |
* The training process uses **Flow Distillation** with Flux.2 as a teacher, forcing the model to learn strong image semantics
|
| 40 |
|
|
|
|
|
|
|
| 41 |
Here is one **4X** upscaled image (two passes):
|
| 42 |
|
| 43 |
|
|
|
|
| 33 |
|
| 34 |
Features:
|
| 35 |
|
| 36 |
+
* Upscaling from **512x512** to **1024x1024** takes **8ms***
|
| 37 |
* The model is trained for **2X** upscaling, but multiple passes can be chained to reach up to **8K** resolution
|
| 38 |
* A full pipeline with Flux generation, upscaling to **8K**, and decoding runs in just **25 seconds** (on RTX 5090)
|
| 39 |
* The training process uses **Flow Distillation** with Flux.2 as a teacher, forcing the model to learn strong image semantics
|
| 40 |
|
| 41 |
+
*On RTX 5090, in latent space, without decoding, see benchmark [here](https://github.com/tensorforger/CTGMWorkshop).
|
| 42 |
+
|
| 43 |
Here is one **4X** upscaled image (two passes):
|
| 44 |
|
| 45 |
|