Hugging Face's logo

SyFeee
/
LTX-2.3-SyFe-Union-Control

Video-to-Video
LTX-2
ltx-video
ltx-2.3
lora
ic-lora
canny
depth
pose
control-video
Model card Files
xet
 Community
SyFeee commited on
Commit
435ff89
·
verified ·
1 Parent(s): 51e8e59

Add files using upload-large-folder tool

Browse files
Files changed (4) hide show
  1. README.md +33 -0
  2. SHA256SUMS +1 -0
  3. runs/union_control_bal/checkpoints/lora_weights_step_05000.safetensors +3 -0
  4. runs/union_control_bal/training_config.yaml +90 -0
README.md ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Lightricks/LTX-2.3
3
+ library_name: ltx
4
+ license: other
5
+ license_name: ltx-2-community-license
6
+ license_link: https://github.com/Lightricks/LTX-2/blob/main/LICENSE
7
+ pipeline_tag: video-to-video
8
+ tags:
9
+ - ltx-video
10
+ - ltx-2.3
11
+ - lora
12
+ - ic-lora
13
+ - canny
14
+ - depth
15
+ - pose
16
+ - control-video
17
+ ---
18
+
19
+ # SyFe LTX-2.3 Union-Control LoRA
20
+
21
+ SyFe-trained full-attention IC-LoRA combining Canny, depth, and pose control for LTX-2.3 22B-dev.
22
+
23
+ ## Checkpoint
24
+
25
+ `union_control_bal` was trained for 5,000 steps at rank 128 on a balanced 19,740-row control corpus: 6,580 Canny, 6,580 depth, and 6,580 pose examples. Training used `video_to_video` conditioning at 1280x704 with reference scale 1.0.
26
+
27
+ The final checkpoint and exact training configuration are under `runs/union_control_bal/`.
28
+
29
+ ## Status
30
+
31
+ Training completed, but the final checkpoint has not received a complete production-quality validation pass. Treat this as a research checkpoint and evaluate each control mode independently before deployment. This is not the downloaded Lightricks Union-Control adapter and does not replace the separately published SyFe pose, depth, and Canny repositories.
32
+
33
+ Use is subject to the LTX-2 community license.
SHA256SUMS ADDED
@@ -0,0 +1 @@
 
 
1
+ 87a6731d273e14430e03041b8a240ba7bdfbf6415df3380e2df48ad81c196423 runs/union_control_bal/checkpoints/lora_weights_step_05000.safetensors
runs/union_control_bal/checkpoints/lora_weights_step_05000.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87a6731d273e14430e03041b8a240ba7bdfbf6415df3380e2df48ad81c196423
3
+ size 1711612856
runs/union_control_bal/training_config.yaml ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ acceleration:
2
+ load_text_encoder_in_8bit: false
3
+ mixed_precision_mode: bf16
4
+ quantization: null
5
+ checkpoints:
6
+ interval: 500
7
+ keep_last_n: 4
8
+ no_resume: false
9
+ precision: bfloat16
10
+ save_training_state: minimal
11
+ data:
12
+ num_dataloader_workers: 2
13
+ preprocessed_data_root: /data0/syfe/lora_output/union_control_bal/preprocessed
14
+ flow_matching:
15
+ timestep_sampling_mode: shifted_logit_normal
16
+ timestep_sampling_params: {}
17
+ hub:
18
+ hub_model_id: null
19
+ push_to_hub: false
20
+ lora:
21
+ alpha: 128
22
+ dropout: 0.0
23
+ rank: 128
24
+ target_modules:
25
+ - to_k
26
+ - to_q
27
+ - to_v
28
+ - to_out.0
29
+ model:
30
+ load_checkpoint: null
31
+ model_path: /data0/pretrain_models/for_video/Lightricks/LTX-2.3/ltx-2.3-22b-dev.safetensors
32
+ text_encoder_path: /data0/pretrain_models/for_text/google/gemma-3-12b-it-qat-q4_0-unquantized
33
+ training_mode: lora
34
+ optimization:
35
+ batch_size: 1
36
+ enable_gradient_checkpointing: true
37
+ gradient_accumulation_steps: 1
38
+ learning_rate: 0.0001
39
+ max_grad_norm: 1.0
40
+ optimizer_type: adamw
41
+ scheduler_params: {}
42
+ scheduler_type: linear
43
+ steps: 5000
44
+ output_dir: /data0/syfe/lora_output/union_control_bal
45
+ seed: 42
46
+ training_strategy:
47
+ first_frame_conditioning_p: 0.1
48
+ name: video_to_video
49
+ reference_latents_dir: reference_latents
50
+ validation:
51
+ frame_rate: 24.0
52
+ generate_audio: false
53
+ guidance_scale: 4.0
54
+ images: null
55
+ include_reference_in_output: false
56
+ inference_steps: 20
57
+ interval: 500
58
+ negative_prompt: no CGI, no animation, no illustration, blurry, distorted, wrong
59
+ pose, extra limbs
60
+ prompts:
61
+ - "\u4E00\u4F4D\u4E2D\u5F0F\u53E4\u88C5\u4EBA\u7269\u5728\u5EAD\u9662\u4E2D\uFF0C\
62
+ \u52A8\u4F5C\u4E0E\u59FF\u6001\u4E25\u683C\u9075\u5FAA\u53C2\u8003\u63A7\u5236\
63
+ \u56FE\u3002\u771F\u4EBA\u7535\u5F71\u611F\uFF0C\u5199\u5B9E\u4E2D\u5F0F\u53E4\
64
+ \u88C5\u5267"
65
+ - "\u4E24\u4F4D\u4EBA\u7269\u5BF9\u8BDD\u4E92\u52A8\uFF0C\u6784\u56FE\u4E0E\u59FF\
66
+ \u6001\u4E25\u683C\u9075\u5FAA\u53C2\u8003\u63A7\u5236\u56FE\u3002\u771F\u4EBA\
67
+ \u7535\u5F71\u611F\uFF0C\u5199\u5B9E\u4E2D\u5F0F\u53E4\u88C5\u5267"
68
+ reference_downscale_factor: 1
69
+ reference_videos:
70
+ - /data0/syfe/lora_output/union_control_v1/val_refs/val_control_a.mp4
71
+ - /data0/syfe/lora_output/union_control_v1/val_refs/val_control_b.mp4
72
+ seed: 42
73
+ skip_initial_validation: true
74
+ stg_blocks:
75
+ - 29
76
+ stg_mode: stg_v
77
+ stg_scale: 1.0
78
+ video_dims: !!python/tuple
79
+ - 1280
80
+ - 704
81
+ - 145
82
+ videos_per_prompt: 1
83
+ wandb:
84
+ enabled: false
85
+ entity: null
86
+ log_validation_videos: true
87
+ project: ltx-2-trainer
88
+ tags:
89
+ - ltx2
90
+ - lora