Update README.md
Browse files
README.md
CHANGED
@@ -10,54 +10,13 @@ tags:
|
|
10 |
pipeline_tag: image-to-video
|
11 |
---
|
12 |
|
13 |
-
Comfyui doesnt natively support these Quants
|
14 |
|
15 |
-
|
16 |
|
17 |
-
|
18 |
|
19 |
-
and change the __init__ function for class LTXVModel(torch.nn.Module):
|
20 |
|
21 |
-
from this:
|
22 |
-
|
23 |
-
|
24 |
-
def __init__(self,
|
25 |
-
in_channels=128,
|
26 |
-
cross_attention_dim=2048,
|
27 |
-
attention_head_dim=64,
|
28 |
-
num_attention_heads=32,
|
29 |
-
|
30 |
-
caption_channels=4096,
|
31 |
-
num_layers=28,
|
32 |
-
|
33 |
-
|
34 |
-
positional_embedding_theta=10000.0,
|
35 |
-
positional_embedding_max_pos=[20, 2048, 2048],
|
36 |
-
causal_temporal_positioning=False,
|
37 |
-
vae_scale_factors=(8, 32, 32),
|
38 |
-
dtype=None, device=None, operations=None, **kwargs):
|
39 |
-
|
40 |
-
|
41 |
-
to this:
|
42 |
-
|
43 |
-
def __init__(self,
|
44 |
-
in_channels=128,
|
45 |
-
cross_attention_dim=4096,
|
46 |
-
attention_head_dim=128,
|
47 |
-
num_attention_heads=32,
|
48 |
-
|
49 |
-
caption_channels=4096,
|
50 |
-
num_layers=48,
|
51 |
-
|
52 |
-
|
53 |
-
positional_embedding_theta=10000.0,
|
54 |
-
positional_embedding_max_pos=[20, 2048, 2048],
|
55 |
-
causal_temporal_positioning=False,
|
56 |
-
vae_scale_factors=(8, 32, 32),
|
57 |
-
dtype=None, device=None, operations=None, **kwargs):
|
58 |
-
|
59 |
-
|
60 |
-
If you want to use the 2b again just revert it and restart. After a restart the 13b model should work with this and the vae i uploaded.
|
61 |
|
62 |
This is a direct GGUF conversion of [Lightricks/ltxv-13b-0.9.7-dev](https://huggingface.co/Lightricks/LTX-Video)
|
63 |
|
|
|
10 |
pipeline_tag: image-to-video
|
11 |
---
|
12 |
|
13 |
+
Comfyui doesnt natively support these Quants in the latest stable version but the nightly/dev build has a fix, which makes the old workaround irrelevant.
|
14 |
|
15 |
+
To update to the dev version just execute the ComfyUI_windows_portable\update\update_comfyui.bat (or equivalent on other install options) and it should load without issues!
|
16 |
|
17 |
+
Also you need to use the right vae. Ive tested it with some old vae for ltxv and it gave massive errors, anyway just use this one here [vae](https://huggingface.co/wsbagnsv1/ltxv-13b-0.9.7-dev-GGUF/blob/main/ltxv-13b-0.9.7-vae-BF16.safetensors)
|
18 |
|
|
|
19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
|
21 |
This is a direct GGUF conversion of [Lightricks/ltxv-13b-0.9.7-dev](https://huggingface.co/Lightricks/LTX-Video)
|
22 |
|