Update 2kto4k description: 4096 px decoding; note both variants support multiple aspect ratios
Browse files
README.md
CHANGED
|
@@ -59,8 +59,9 @@ Two variants are released for each diffusers-style backbone:
|
|
| 59 |
- **`2k`** β trained at 2048px, used as a 4Γ decoder (512 LDM β 2048 px), or as
|
| 60 |
an 8Γ decoder for the Scale-RAE backbone (256 β 2048).
|
| 61 |
- **`2kto4k`** β trained with multi-resolution data bucketing 2048β3840 and an
|
| 62 |
-
SD3-style dynamic shift; designed for 1024 LDM β 4K (
|
| 63 |
-
|
|
|
|
| 64 |
|
| 65 |
| Path | Backbone (encoder side) | SR factor | Variant |
|
| 66 |
|---------------------------------------------------------------|--------------------------------------------|-----------|-----------|
|
|
|
|
| 59 |
- **`2k`** β trained at 2048px, used as a 4Γ decoder (512 LDM β 2048 px), or as
|
| 60 |
an 8Γ decoder for the Scale-RAE backbone (256 β 2048).
|
| 61 |
- **`2kto4k`** β trained with multi-resolution data bucketing 2048β3840 and an
|
| 62 |
+
SD3-style dynamic shift; designed for 1024 LDM β 4K (4096 px) decoding.
|
| 63 |
+
|
| 64 |
+
Both checkpoint variants support multiple aspect ratios.
|
| 65 |
|
| 66 |
| Path | Backbone (encoder side) | SR factor | Variant |
|
| 67 |
|---------------------------------------------------------------|--------------------------------------------|-----------|-----------|
|