Add comprehensive LTX-Video support (LTXVideo, LTXVideoI2V, LTX2Video, LTX2VideoI2V) by Copilot · Pull Request #35 · nodetool-ai/nodetool-huggingface

Copilot · 2026-04-28T09:07:43Z

Summary

Implements comprehensive support for LTX-Video (Lightricks) models, including the newest LTX-2 architecture, with both text-to-video and image-to-video pipelines.

Changes

src/nodetool/nodes/huggingface/text_to_video.py

LTXVideo — Text-to-video using LTXPipeline with three model variants:
- Lightricks/LTX-Video-0.9.5 (recommended, latest)
- Lightricks/LTX-Video-0.9.1
- Lightricks/LTX-Video (original)
LTX2Video — Text-to-video using LTX2Pipeline for Lightricks/LTX-2 (newest architecture with Gemma3 text encoder, long prompt support up to 1024 tokens)

src/nodetool/nodes/huggingface/image_to_video.py

LTXVideoI2V — Image-to-video using LTXImageToVideoPipeline with the same three model variants as above
LTX2VideoI2V — Image-to-video using LTX2ImageToVideoPipeline for Lightricks/LTX-2

Design Details

All nodes follow the established patterns in this repository:

CPU offload via enable_model_cpu_offload() to reduce VRAM usage
VAE slicing and tiling options for memory efficiency
Progress callbacks for step-by-step UI updates
Reproducible generation via seed control
get_basic_fields(), get_recommended_models(), get_title() implemented
required_inputs() for image-to-video nodes

LTX-Video nodes use available_torch_dtype() (bfloat16/float16 depending on hardware). LTX-2 nodes use torch.bfloat16 explicitly as required by the model.

…, LTX2VideoI2V nodes Agent-Logs-Url: https://github.com/nodetool-ai/nodetool-huggingface/sessions/3a47e6ee-3e27-42ae-bb22-f77a93bee7b9 Co-authored-by: georgi <19498+georgi@users.noreply.github.com>

Copilot

Pull request overview

Adds new HuggingFace pipeline nodes to generate videos with Lightricks LTX-Video and LTX-2 models, covering both text-to-video and image-to-video workflows.

Changes:

Add LTXVideo and LTX2Video text-to-video nodes using diffusers LTX/LTX2 pipelines.
Add LTXVideoI2V and LTX2VideoI2V image-to-video nodes using diffusers LTX/LTX2 image2video pipelines.
Introduce model-variant selection, seed control, CPU offload, and VAE slicing/tiling options for these nodes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 11 comments.

File	Description
src/nodetool/nodes/huggingface/text_to_video.py	Adds LTX-Video and LTX-2 text-to-video nodes and related diffusers pipeline imports.
src/nodetool/nodes/huggingface/image_to_video.py	Adds LTX-Video and LTX-2 image-to-video nodes and related diffusers pipeline imports.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-28T09:43:17Z

+    frame_rate: int = Field(
+        default=25,
+        description="Frame rate for the output video file.",
+        ge=1,
+        le=60,
+    )


This node introduces frame_rate for output FPS, but other image-to-video nodes in this file use the fps field name. For consistency across nodes (and to reduce UI/API special-casing), consider renaming this field to fps and using it consistently for video_from_frames output metadata; if the pipeline requires frame_rate, pass frame_rate=self.fps internally.

Copilot · 2026-04-28T09:43:18Z

+    frame_rate: int = Field(
+        default=25,
+        description="Frame rate for the output video file.",
+        ge=1,
+        le=60,
+    )


This node uses frame_rate for output FPS, but the rest of the video nodes in this module use the fps field name (and pass fps=... to video_from_frames/video_from_numpy). Using a different name makes the API/UI inconsistent and increases maintenance burden. Consider renaming this field to fps (and, if the diffusers pipeline needs frame_rate, pass frame_rate=self.fps internally).

Suggested change

frame_rate: int = Field(

default=25,

description="Frame rate for the output video file.",

ge=1,

le=60,

)

fps: int = Field(

default=25,

alias="frame_rate",

description="Frame rate for the output video file.",

ge=1,

le=60,

)

@property

def frame_rate(self) -> int:

return self.fps

Copilot · 2026-04-28T09:43:18Z

+    frame_rate: float = Field(
+        default=24.0,
+        description="Frame rate for the output video file.",
+        ge=1.0,
+        le=60.0,


frame_rate is defined as a float but video_from_frames requires an integer FPS, and the implementation truncates via int(self.frame_rate). This can silently change user input (e.g., 23.976 → 23). Prefer an fps: int field (consistent with other nodes) or validate/round explicitly so the stored metadata matches the requested frame rate.

Suggested change

frame_rate: float = Field(

default=24.0,

description="Frame rate for the output video file.",

ge=1.0,

le=60.0,

frame_rate: int = Field(

default=24,

description="Frame rate for the output video file in integer FPS.",

ge=1,

le=60,

Copilot · 2026-04-28T09:43:18Z

+    num_frames: int = Field(
+        default=161,
+        description="Total frames in the output video. Must be 8n+1 (e.g. 65, 97, 129, 161). More frames = longer video.",
+        ge=9,
+        le=257,
+    )


num_frames is documented as requiring the form 8n+1, but the current validation only enforces a range (ge/le). This allows invalid values (e.g., 10) that contradict the docs and may cause runtime errors. Add explicit validation (e.g., enforce (num_frames - 1) % 8 == 0) or constrain the UI/field to only valid values.

Copilot · 2026-04-28T09:43:19Z

+    frame_rate: float = Field(
+        default=24.0,
+        description="Frame rate for the output video file.",
+        ge=1.0,
+        le=60.0,


frame_rate is a float but the implementation later truncates to an int for encoding (fps = int(self.frame_rate)), which can silently change user intent (e.g., 23.976 → 23). Prefer an fps: int field (matching other nodes) or validate/round explicitly and reflect the effective FPS in the output metadata.

Suggested change

frame_rate: float = Field(

default=24.0,

description="Frame rate for the output video file.",

ge=1.0,

le=60.0,

frame_rate: int = Field(

default=24,

description="Integer frame rate (FPS) for the output video file.",

ge=1,

le=60,

Copilot · 2026-04-28T09:43:19Z

+class LTXVideoI2V(HuggingFacePipelineNode):
+    """
+    Animates a static image into a video using LTX-Video image-to-video diffusion models.
+    video, generation, AI, image-to-video, diffusion, LTX, Lightricks, animation
+
+    Use cases:
+    - Animate photographs or artwork into dynamic video clips
+    - Create motion-driven video from a single input image with text guidance
+    - Generate content for social media or creative projects from still images
+    - Bring product images or illustrations to life
+    - Produce smooth animated sequences with fine temporal control
+
+    **Note:** LTX-Video-0.9.5 is recommended for best quality.
+    """


New LTX image-to-video nodes are added without a lightweight import/smoke test. Consider adding an import-only test to ensure these nodes (and the diffusers.pipelines.ltx* imports) remain available with the supported dependency versions, similar to other HuggingFace node smoke tests.

Copilot · 2026-04-28T09:43:19Z

+    num_frames: int = Field(
+        default=161,
+        description="Total frames in the output video. Must be 8n+1 (e.g. 65, 97, 129, 161). More frames = longer video.",
+        ge=9,
+        le=257,
+    )


num_frames is documented as requiring the form 8n+1, but the current validation only enforces a range (ge/le). This allows values like 10 or 100 that will violate the documented requirement and may fail at runtime. Add explicit validation (e.g., a Pydantic validator enforcing (num_frames - 1) % 8 == 0) or adjust the field constraints/UI to only permit valid values.

Copilot · 2026-04-28T09:43:20Z

+        default=1024,
+        description="Maximum prompt encoding length. LTX-2 supports long prompts up to 1024 tokens.",
+        ge=64,
+        le=2048,


The description says LTX-2 supports prompts up to 1024 tokens, but the field allows values up to 2048 (le=2048). Either update the description to match the actual supported limit or tighten the constraint to avoid allowing values that the model/pipeline may reject.

Suggested change

le=2048,

le=1024,

Copilot · 2026-04-28T09:43:20Z

+    - Generate videos with improved temporal consistency and visual fidelity
+    - Produce cinematic content with detailed motion and scene description
+    - Build next-generation AI video generation workflows
+    - Create visual content with support for audio output


The docstring claims “support for audio output”, but the node only returns frames/video and doesn’t expose any audio-related parameters or outputs. If audio isn’t actually supported here, please remove/adjust this bullet to avoid misleading users; if it is supported, the node should surface the audio output in its return type/metadata.

Suggested change

- Create visual content with support for audio output

- Create visual content for video-first generation workflows

Copilot · 2026-04-28T09:43:20Z

+class LTXVideo(HuggingFacePipelineNode):
+    """
+    Generates high-quality videos from text prompts using LTX-Video diffusion models.
+    video, generation, AI, text-to-video, diffusion, LTX, Lightricks, cinematic
+
+    Use cases:
+    - Create smooth, high-fidelity videos from detailed text descriptions
+    - Generate cinematic and artistic video content
+    - Produce short animated clips for social media and creative projects
+    - Build AI-driven video generation pipelines
+    - Create visual content with fine-grained temporal control
+
+    **Note:** LTX-Video-0.9.5 is recommended for best quality; older versions are also supported.
+    """


New LTX nodes are added, but there’s no accompanying lightweight test to ensure these modules/nodes can be imported in the supported dependency set (e.g., that the diffusers.pipelines.ltx* entrypoints exist). Consider adding an import-only smoke test similar to existing HuggingFace node tests to catch dependency/version regressions early.

Copilot AI assigned Copilot and georgi Apr 28, 2026

Copilot created this pull request from a session on behalf of georgi April 28, 2026 09:08 View session

Copilot AI requested a review from georgi April 28, 2026 09:08

Copilot finished work on behalf of georgi April 28, 2026 09:08

georgi marked this pull request as ready for review April 28, 2026 09:37

georgi requested a review from Copilot April 28, 2026 09:37

Copilot started reviewing on behalf of georgi April 28, 2026 09:38 View session

Copilot AI reviewed Apr 28, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add comprehensive LTX-Video support (LTXVideo, LTXVideoI2V, LTX2Video, LTX2VideoI2V)#35

Add comprehensive LTX-Video support (LTXVideo, LTXVideoI2V, LTX2Video, LTX2VideoI2V)#35
Copilot wants to merge 1 commit into
mainfrom
copilot/implement-comprehensive-support-ltx

Copilot AI commented Apr 28, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	- Create visual content with support for audio output
	- Create visual content for video-first generation workflows

Conversation

Copilot AI commented Apr 28, 2026

Summary

Changes

Design Details

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants