Update README.md
Browse files
README.md
CHANGED
|
@@ -3,3 +3,17 @@ license: other
|
|
| 3 |
license_name: ideogram-4-non-commercial
|
| 4 |
license_link: https://huggingface.co/ideogram-ai/ideogram-4-fp8/blob/main/LICENSE.md
|
| 5 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
license_name: ideogram-4-non-commercial
|
| 4 |
license_link: https://huggingface.co/ideogram-ai/ideogram-4-fp8/blob/main/LICENSE.md
|
| 5 |
---
|
| 6 |
+
|
| 7 |
+
Repackages of Fal's [Ideogram4-Fast](https://huggingface.co/fal/ideogram-v4-fast) and [Ideogram4-Instant](https://huggingface.co/fal/ideogram-v4-instant) to ComfyUI compatible files, with int8 and nvfp4 conversion done with [Silveroxides/convert_to_quant](https://github.com/silveroxides/convert_to_quant), targeting only attention and MLP tensors.
|
| 8 |
+
|
| 9 |
+
Officially recommended parameters:
|
| 10 |
+
- for Fast: 20 steps, CFG 1
|
| 11 |
+
- for Instant: 8 steps, CFG 1
|
| 12 |
+
|
| 13 |
+
Reported gen speed on a 5090 at 1536x1536:
|
| 14 |
+
- bf16: 0.85 it/s
|
| 15 |
+
- int8: 1.7 it/s
|
| 16 |
+
- nvfp4: 1.9 it/s
|
| 17 |
+
|
| 18 |
+
Note that Fast has QAD (Quantization Aware Distillation) and can be better than its larger variants, especially on long texts.
|
| 19 |
+
Fal's readme of Instant says it doesn't have QAD but it's just as good as the larger ones too.
|