Cseti commited on
Commit
e02c4b1
·
1 Parent(s): d14def2

crossview-warp workflow uploaded

Browse files
README.md CHANGED
@@ -23,6 +23,7 @@ Drag-and-drop any `.json` file directly into ComfyUI to load the workflow.
23
 
24
  | Workflow | Description | Type | Updated |
25
  | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | --------- | ---------- |
 
26
  | [IC-LoRA CrossView (v1 pilot)](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-crossview-v1-pilot/README.md) | V2V new-camera-angle IC-LoRA — re-render a clip from a prompt-described virtual second camera (pilot, step 13.7k) | Video gen | 2026-07-11 |
27
  | [IC-LoRA Cameraman v2](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-cameraman-v2/README.md) | I2V with IC-LoRA camera motion guidance (Cameraman v2, step 14k) — dev + distilled LoRA + 2x upscaler | Video gen | 2026-06-24 |
28
  | [Arcane Jinx Character LoRA](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/t2v-arcane-jinx-lora/README.md) | T2V with audio using a custom-trained Arcane Jinx character LoRA on LTX 2.3 — 22B | Video gen | 2026-05-06 |
 
23
 
24
  | Workflow | Description | Type | Updated |
25
  | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | --------- | ---------- |
26
+ | [IC-LoRA CrossView Warp (v0.9)](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-crossview-warp/README.md) | V2V new-camera-angle IC-LoRA — set the virtual second camera by numbers (azimuth/elevation/distance) via a depth-warp node (PoC, step 18k) | Video gen | 2026-07-24 |
27
  | [IC-LoRA CrossView (v1 pilot)](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-crossview-v1-pilot/README.md) | V2V new-camera-angle IC-LoRA — re-render a clip from a prompt-described virtual second camera (pilot, step 13.7k) | Video gen | 2026-07-11 |
28
  | [IC-LoRA Cameraman v2](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-cameraman-v2/README.md) | I2V with IC-LoRA camera motion guidance (Cameraman v2, step 14k) — dev + distilled LoRA + 2x upscaler | Video gen | 2026-06-24 |
29
  | [Arcane Jinx Character LoRA](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/t2v-arcane-jinx-lora/README.md) | T2V with audio using a custom-trained Arcane Jinx character LoRA on LTX 2.3 — 22B | Video gen | 2026-05-06 |
ltx/2.3/ic-lora-crossview-warp/README.md ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # LTX Video 2.3 — IC-LoRA CrossView Warp (v0.9 PoC) WORKFLOW
2
+
3
+ **Model:** LTX Video 2.3 — 22B Dev (FP8) + distilled speed LoRA
4
+ **Type:** Video-to-Video
5
+
6
+ Video-to-Video workflow using LTX Video 2.3 with the [CrossView Warp IC-LoRA](https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Warp) — it re-renders an input clip from a new virtual camera angle, like a second camera on the same take. Unlike the [CrossView Prompt](https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-crossview-v1-pilot/README.md) variant (which picks the angle from a text phrase), here the angle is set as **numbers** — azimuth / elevation / distance — on the [CrossViewWarp node](https://github.com/cseti007/ComfyUI-CrossViewWarp). That node builds a Depth-Anything-V2 depth-warp of the input, and the graph feeds two IC-LoRA reference guides (the warp for geometry, the source clip for identity). Runs on the 22B **dev** transformer with the distilled speed LoRA and a 2x latent spatial upscaler.
7
+
8
+ ---
9
+
10
+ ## Preview
11
+
12
+ <video src="https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/resolve/main/ltx/2.3/ic-lora-crossview-warp/media/preview.mp4" controls width="720"></video>
13
+
14
+ Reference input on the left, new-angle generated output on the right.
15
+
16
+ <!-- Replace YOUTUBE_VIDEO_ID with your video ID after uploading -->
17
+ <!-- [![Watch on YouTube — click to open](https://img.youtube.com/vi/YOUTUBE_VIDEO_ID/maxresdefault.jpg)](https://www.youtube.com/watch?v=YOUTUBE_VIDEO_ID) -->
18
+ <!-- *Click to open on YouTube* -->
19
+
20
+ ---
21
+
22
+ ## Requirements
23
+
24
+ - **ComfyUI:** recent stable build
25
+ - **Model:** `ltx-2.3-22b-dev_transformer_only_fp8_scaled.safetensors` — [Kijai/LTX2.3_comfy](https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/diffusion_models/ltx-2.3-22b-dev_transformer_only_fp8_scaled.safetensors)
26
+ - **Distilled speed LoRA:** `ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors` — [Kijai/LTX2.3_comfy](https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/loras/ltx-2.3-22b-distilled-1.1_lora-dynamic_fro09_avg_rank_111_bf16.safetensors)
27
+ - **IC-LoRA:** `LTX2.3-22B_IC-LoRA-CrossView-Warp_v0.9_18000.safetensors` — [Cseti/LTX2.3-22B_IC-LoRA-CrossView-Warp](https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Warp) (place under `models/loras/LTX2/`)
28
+ - **Video VAE:** `LTX23_video_vae_bf16.safetensors` — [Kijai/LTX2.3_comfy](https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/vae/LTX23_video_vae_bf16.safetensors)
29
+ - **Audio VAE:** `LTX23_audio_vae_bf16.safetensors` — [Kijai/LTX2.3_comfy](https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/vae/LTX23_audio_vae_bf16.safetensors)
30
+ - **Preview VAE:** `taeltx2_3.safetensors` — [Kijai/LTX2.3_comfy](https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/vae/taeltx2_3.safetensors)
31
+ - **Text encoder:** `gemma_3_12B_it_fp8_scaled.safetensors` — [Comfy-Org/ltx-2](https://huggingface.co/Comfy-Org/ltx-2/blob/main/split_files/text_encoders/gemma_3_12B_it_fp8_scaled.safetensors)
32
+ - **Text projection:** `ltx-2.3_text_projection_bf16.safetensors` — [Kijai/LTX2.3_comfy](https://huggingface.co/Kijai/LTX2.3_comfy/blob/main/text_encoders/ltx-2.3_text_projection_bf16.safetensors)
33
+ - **Spatial upscaler:** `ltx-2.3-spatial-upscaler-x2-1.1.safetensors` — [Lightricks/LTX-2.3](https://huggingface.co/Lightricks/LTX-2.3/blob/main/ltx-2.3-spatial-upscaler-x2-1.1.safetensors)
34
+ - **Depth model:** `depth_anything_v2_vitl_fp32.safetensors` — auto-downloaded by the `DownloadAndLoadDepthAnythingV2Model` node
35
+ - **Custom nodes:**
36
+ - [ComfyUI-CrossViewWarp](https://github.com/cseti007/ComfyUI-CrossViewWarp) — the CrossView Warp node (companion to the IC-LoRA)
37
+ - [ComfyUI-LTXVideo](https://github.com/Lightricks/ComfyUI-LTXVideo)
38
+ - [ComfyUI-KJNodes](https://github.com/kijai/ComfyUI-KJNodes)
39
+ - [ComfyUI-VideoHelperSuite](https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite)
40
+ - [ComfyUI-DepthAnythingV2](https://github.com/kijai/ComfyUI-DepthAnythingV2)
41
+ - [rgthree-comfy](https://github.com/rgthree/rgthree-comfy)
42
+ - [RES4LYF](https://github.com/ClownsharkBatwing/RES4LYF)
43
+ - [ComfyUI_Fill-Nodes](https://github.com/filliptm/ComfyUI_Fill-Nodes)
44
+
45
+ ---
46
+
47
+ ## Notes
48
+
49
+ - **Input:** a reference video clip is loaded via the `Load Video` node. It is fed to the **CrossViewWarp node** (frames + Depth-Anything-V2 depth) which outputs the depth-warp; both the warp and the source clip go to their own `LTXAddVideoICLoRAGuide`.
50
+ - **Angle:** set `azimuth` / `elevation` / `distance` on the CrossViewWarp node, or drag the marker on its built-in 3D orbit picker. Stay in the green zone — azimuth up to ±45°, elevation +30°/−15° (yellow up to ±65° works with degrading quality). Full angle-range table is on the [LoRA card](https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Warp).
51
+ - **IC-LoRA strength `1.3`** (1.0–1.15 for cars and content the base model has strong priors about); distilled speed LoRA at `0.6`.
52
+ - **Both IC-LoRA guides use `latent_downscale_factor = 1`** — see the LoRA card for why (a training-time downscale mismatch).
53
+ - For `distance > 1`: describe the newly revealed content in the text prompt — the warp can't know what's outside the original frame.
54
+ - **Two-pass:** an 8-step base pass (~960x544, 97 frames @ 24fps, with audio) followed by a 2x latent spatial upscale pass. The upscale pass reuses the source clip through the IC-LoRA (without the warp condition) for better likeness.
55
+ - This is a **proof-of-concept (v0.9)** checkpoint (step 18,000) — the model steers the viewpoint rather than reprojecting it, so you often get a smaller rotation than requested. Read the LoRA card's limitations before expecting exact angles.
56
+
57
+ ---
58
+
59
+ ## Changelog
60
+
61
+ - `2026-07-24` — Initial upload
62
+
63
+ ---
64
+
65
+ ## Training
66
+
67
+ The IC-LoRA was trained on [RunPod cloud GPUs](https://runpod.io?ref=vpp0cion) (RTX PRO 6000 Blackwell). Full training details are on the [LoRA card](https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Warp).
68
+
69
+ ---
70
+
71
+ ## Support
72
+
73
+ Producing and sharing this kind of open-source work requires renting cloud GPUs, which gets expensive quickly. If you find it useful and would like me to keep contributing, your support is very much appreciated:
74
+
75
+ [![Ko-fi](https://img.shields.io/badge/Ko--fi-Support-FF5E5B?style=for-the-badge&logo=ko-fi&logoColor=white)](https://ko-fi.com/chetyart) [![Liberapay](https://img.shields.io/badge/Liberapay-Donate-F6C915?style=for-the-badge&logo=liberapay&logoColor=black)](https://liberapay.com/chetyart/donate)
ltx/2.3/ic-lora-crossview-warp/ltx2.3-ic-lora-crossview-warp.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2917fdab81eb8e671674ed7626925e24471a50e7153fed29ca91643c45c998fd
3
+ size 121504
ltx/2.3/ic-lora-crossview-warp/media/preview.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52efd50d1fdbd3487ce72481f6d5838a69c6d5d17ee5e797b9b945a09b9f823c
3
+ size 1639542
ltx/2.3/ic-lora-crossview-warp/media/preview_thumb.png ADDED

Git LFS Details

  • SHA256: 8a00fe613e1d2a215cbfffcaca7830fd1114a830353b77d90723bc9fe87e0ed2
  • Pointer size: 131 Bytes
  • Size of remote file: 804 kB