Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt · Hugging Face

Source: https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#ltx-video-23-22b–ic-lora-crossview-prompt-v09LTX-Video 2.3 22B — IC-LoRA: CrossView Prompt v0.9

A fine-tuned In-Context LoRA (IC-LoRA) adapter for LTX-Video 2.3 (22B) that acts as avirtual second camera: give it a reference video and a short camera-angle prompt, and it re-renders the same scene from the requested new viewpoint keeping the subject and content, changing where the camera stands.

**v0.9 — proof-of-concept.**Trained on synthetic multi-view data; it generalizes to real footage but has clear limits (see Limitations). Feedback welcome.

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#example-outputsExample outputs

Each video shows thereference(top) and thegenerated new camera view(bottom) for the given prompt.

Promptcrossview. new camera angle: to the right, lower, closer. Promptcrossview. new camera angle: to the right, lower, further. Promptcrossview. new camera angle: to the right, lower, closer. Promptcrossview. new camera angle: to the left, higher, further. Promptcrossview. new camera angle: to the left, higher, further. Promptcrossview. new camera angle: to the left, higher, further. Promptcrossview. new camera angle: to the left, higher, closer. Promptcrossview. new camera angle: to the left, higher, further.

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#usage-comfyuiUsage (ComfyUI)

I tested this LoRA only inComfyUI, in a video-to-video (IC-LoRA) workflow. An example workflow is here:https://huggingface.co/datasets/Cseti/ComfyUI-Workflows/blob/main/ltx/2.3/ic-lora-crossview-v1-pilot/README.md

How it works:

• LoadLTX2\.3\-22B\_IC\-LoRA\-CrossView\-Prompt\_v0\.9\_13700\.safetensorsas the LoRA.
• Provide areference video— the scene you want to re-shoot from a new angle.
• Provide acamera-angle prompt(see the vocabulary below). No starting image is needed

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#prompt-vocabulary-importantPrompt vocabulary (important)

Unlike a free-text LoRA, this model was trained on afixed, discrete camera vocabulary. Every prompt must start with the triggercrossview\.followed by the template:

crossview. new camera angle: {azimuth}, {elevation}, {distance}.

AxisAllowed phrasesazimuth(orbit around the subject)same angle·slightly to the left·slightly to the right·to the left·to the right·far to the left·far to the rightelevation(camera height)lower·same height·higherdistance(to the subject)closer·same distance·further

• left / right= the new camera moves to that sidearound the subject.
• higher= the camera looks down from above;lower= looks up from below.
• closer / further= the camera’s distance to the subject.

All 63 valid combinations are listed incaptions\_all\_63\.txt. Use these exact phrases — the model learned this vocabulary specifically, so synonyms (“45 degrees left”, “slightly leftward”) work less reliably.

Example prompts:

crossview. new camera angle: to the right, lower, closer.
crossview. new camera angle: to the left, higher, further.
crossview. new camera angle: same angle, same height, closer.

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#tipsTips

• Angle size & chaining:the model works most reliably onsmall, single-step angle changes. For a larger viewpoint shift,chain several small steps— feed the generated view back in as the new reference and apply another small angle.
• **Full prompt list:**every prompt used to train this model is incaptions\_all\_63\.txt— use these exact phrases.
• Distilled model:the LoRA was trained on the full (non-distilled) LTX-2.3. On distilled few-step workflows its effect is weaker — try aLoRA strength of 1.2–1.5, and/or run it in the first (non-distilled) pass.

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#training-detailsTraining Details

This IC-LoRA was trained onRunPod cloud GPUs(NVIDIA RTX PRO 6000 Blackwell, 96 GB).

ParameterValueBase modelLTX-Video 2.3 (22B)Training frameworkltx-trainer (Lightricks)Training strategyIC-LoRA (flexible, reference conditioning)Released checkpointstep 13,700 (12k linear + 1.7k warm-start continuation)LoRA rank / alpha16 / 16Target modulesattn1, attn2 (to_k/q/v/out) —attention onlyOptimizerAdamW — 2e-4 linear decay (main), 1e-4 constant (continuation)Timestep samplinguniform sigma**[0.4, 1.0]**Mixed precisionbf16Batch size1 (gradient checkpointing enabled)Conditioningreference p=1.0 + first_frame p=0.2Training dataset294 pairsResolution768x768x81 @ 15fps

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#datasetDataset

Trained onSynCamVideo(KlingTeam, Apache-2.0) — a synthetic multi-camera dataset rendered in Unreal Engine 5, with 10 static cameras per scene sampled on a hemisphere around the subject. 294 curated reference/target camera pairs, balanced across the caption vocabulary:

Azimuth binPairssame angle42slightly to the left / right42 / 42to the left / right42 / 42far to the left / right42 / 42 Elevation and distance are mixed roughly evenly within each azimuth bin. Captions are camera-delta only (no scene description).

The exact curated pairs, captions and clips used to train this LoRA are released as a companion dataset:CrossView Prompt Dataset.

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#limitationsLimitations

• Viewpoint range:the training cameras span afrontal sector (~±60° azimuth max)— “view from behind” is out of range.
• **Distilled model:**weaker on distilled few-step models (see Tips).

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#licenseLicense

This LoRA is shared under theApache License 2.0. It was trained entirely on theSynCamVideodataset, which is itself Apache-2.0 licensed, so the training data places no additional restrictions on this adapter and it can be released under the same permissive terms.

Note: using this LoRA requires theLTX-Video 2.3base model, which is governed by its own license — please review Lightricks’ terms for the base weights separately.

https://huggingface.co/Cseti/LTX2.3-22B_IC-LoRA-CrossView-Prompt#supportSupport

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: