IC-LoRA Adapters

This page documents key IC-LoRAs trained by LTX. IC-LoRAs use in-context conditioning to perform specialized tasks on video, from structural control and motion guidance to VFX, video restoration, and creative transforms. Each adapter is trained for a specific task.

For an introduction to how IC-LoRAs work, see the IC-LoRA Guide.

For a versioned directory, see the LTX IC-LoRA catalog below. To browse adapters on Hugging Face, see the LTX-2.5 Creative Lab and LTX-2.3 Creative Lab collections.

Control

IC-LoRAs that guide generation using structural conditioning signals such as depth maps, edge contours, pose skeletons, or motion trajectories, as well as reference sheets that maintain character and object consistency.

Union Control

A single IC-LoRA that handles multiple control types (depth, canny, and pose) in one checkpoint.

Model: LTX-2.3-22b-IC-LoRA-Union-Control

Use cases:

• Apply depth, canny, and pose control from a single model without swapping IC-LoRA checkpoints between runs
• Simplify ComfyUI workflows by replacing three separate IC-LoRA models with one unified control adapter
• Reduce VRAM footprint when working with multiple control types by keeping a single IC-LoRA in memory

Workflow: ComfyUI Union Control workflow

Motion Control

Guides object motion using sparse spline-based trajectories rendered as trails of circles. Supports single or multiple simultaneous motion paths for directing object movement within a scene.

Model: LTX-2.3-22b-IC-LoRA-Motion-Track-Control

Use cases:

• Direct object paths in product showcases, character scenes, or nature shots by drawing spline trajectories
• Create intentional, controlled object motion that is more precise than description via text
• Guide multiple objects simultaneously along independent paths within a single generation

Key nodes: This IC-LoRA uses two dedicated ComfyUI nodes for defining motion paths:

• LTX Draw Tracks — Visual canvas for drawing spline-based motion trajectories
• LTX Sparse Track Editor — fine-tunes keypoint positions and timing and converts sparse keypoints into interpolated per-frame coordinate paths

See Sparse Track Conditioning in the IC-LoRA Guide for best practices on preparing motion paths.

Workflow: ComfyUI Motion Control workflow

Ingredients

Conditions video generation on a reference sheet: a single composite image showing characters, props, and locations laid out on a black background. Generated videos keep those elements visually consistent. Requires the reference sheet as a static video and a two-part prompt (Reference sheet: <panels> / Generated video: <action>); the model renders a clip whose characters, props, and setting match the sheet.

Models: LTX-2.3-22b-IC-LoRA-Ingredients and LTX-2.5-22b-IC-LoRA-Ingredients

Use cases:

• Maintain character consistency (face, costume, body) across multiple generated clips using a shared reference sheet
• Lock props and location details to a predefined visual inventory for scene-to-scene continuity
• Generate action sequences where recurring characters stay on-model without per-frame manual control

LTX-2.5 compatibility: An official LTX-2.5 adapter and dedicated single-stage workflow are available. The workflow currently loads the LTX-2.3 Ingredients checkpoint, while the LTX-2.5 adapter card specifies a separate LTX-2.5 checkpoint. Follow the exact checkpoint required by the workflow or adapter card you are using; do not substitute one for the other.

---

Utility

IC-LoRAs that enhance or extend existing video by filling in missing regions or improving resolution and detail while maintaining the subject or composition.

Pixel Spatial Upscaler

Creatively upscales low-resolution video to 2× or 4× the input resolution, synthesizing fine spatial detail rather than interpolating existing pixels. Separate 2× and 4× variants are provided. The recommended flow is to generate a draft at a low base resolution (around 280p) to lock composition and motion, then run the upscaler to produce a high-resolution result with synthesized detail.

Model: LTX-2.3-22b-IC-LoRA-Pixel-Spatial-Upscaler (2× and 4× variants). An LTX-2.5 adapter is also available, currently in a 2× variant only.

Use cases:

• Produce a high-resolution final render from a fast low-resolution draft, adding synthesized detail
• Choose the 2× variant for moderate upscaling or the 4× variant for larger jumps in resolution
• Add fine spatial detail as a creative step in a generation flow rather than as a faithful refiner

LTX-2.5: The official LTX-2.5 adapter card covers the 2× variant. Follow that card for its supported setup; no dedicated LTX-2.5 workflow is linked here.

For a draft-then-upscale workflow, see Upscaling.

Inpainting / Outpainting

Extends the canvas of existing video (outpainting) or fills masked regions within video frames (inpainting), generating new content that blends seamlessly with the surrounding scene. Uses mask-aware preprocessing and Laplacian pyramid blending at both generation stages for clean integration.

Model: LTX-2.3-22b-IC-LoRA-In-Outpainting

Use cases:

• Extend video beyond its original frame boundaries to create wider or taller compositions
• Fill masked regions of video with generated content that matches the surrounding scene
• Change aspect ratios of existing footage by generating the missing areas

For setup, workflow details, and node documentation, see the In-Outpainting guide.

---

Specialized workflows

These adapters use task-specific workflows and versioned model stacks.

SDR to HDR

Converts SDR video to HDR, producing an EXR sequence plus a BT.2020/HLG master. The current ComfyUI workflow and native Python pipeline default to scene-linear ACEScg EXR. This is a reconstruction workflow applied to existing video, not preservation of an existing HDR source.

Model: LTX-2.5 22B IC-LoRA SDR to HDR

LTX-2.5 files: ltx-2.5-22b-ic-lora-sdr-to-hdr-1.0.safetensors and ltx-2.5-22b-ic-lora-sdr-to-hdr-scene-emb.safetensors

Compatibility: LTX-2.5 ComfyUI and native Python use the dedicated single-stage pipeline, fixed scene embeddings, and no runtime text encoder. The legacy LTX-2.3 HDR adapter uses a separate LogC3 model stack. Keep the versions separate.

For setup, usage, and DaVinci Resolve import instructions, see SDR to HDR.

Alpha Gen (Beta)

Generates a frame-aligned grayscale alpha matte from an RGB source video. The model identifies what it interprets as the main subject. When used as an alpha channel, white keeps the model-selected area fully opaque, black makes the unselected area fully transparent, and gray produces partial opacity.

Model: LTX-2.5 22B IC-LoRA Alpha Gen

LTX-2.5 file: ltx-2.5-22b-ic-lora-alpha-gen-0.9.safetensors

Compatibility: LTX-2.5 ComfyUI and native Python. Use an empty prompt and adapter strength 1.0. The source is limited to 1920×1088. Use 121 frames or fewer for the most predictable results; valid 8n+1 frame counts above 121 and up to 145 may work but can be less reliable, and longer clips are unsupported. Prompt- or mask-based subject selection is not supported.

For setup, usage, and compositing instructions, see Alpha Gen.

Layout to Render

Turns a clay viewport animation or blocky playblast into a finished shot while keeping the layout’s camera path, object placement, and motion. An art-directed image made from the layout’s first frame supplies the palette, lighting, materials, and detail.

Model: LTX-2.5 22B IC-LoRA Layout to Render

LTX-2.5 file: ltx-2.5-22b-ic-lora-layout-to-render-1.0.safetensors

Compatibility: LTX-2.5 ComfyUI and native Python. The model card specifies adapter strength 1.0 with the LTX-2.5 distilled stack and 2× latent spatial upscaler. Follow the model card and guide for current input-geometry recommendations.

For setup and usage, see Layout to Render.

Refine & Restore

Adds synthesized detail and clarity to existing footage through the Tiled Fusion Upscale workflow. It is a generative refinement path rather than pixel-accurate restoration.

Workflow model: Refine Details

Compatibility: LTX-2.5 ComfyUI. The Tiled Fusion Upscale workflow selects the Refine Details adapter. The separately published Restore adapter is not selected by that graph. Do not substitute Restore or the Pixel Spatial Upscaler unless the model card and workflow explicitly call for it.

For setup and usage, see Refine & Restore.

Dub-It (Beta)

Re-generates speech in video, producing lip-synced output with new dialogue while preserving the speaker’s visual appearance and vocal identity. Works with live-action and animated subjects. Dub-It is a video-to-video dubbing tool focused on speech replacement — it can also be used for rephrasing dialogue in the original language.

Model: LTX-2.3-22b-IC-LoRA-DubIt (ltx-2.3-22b-ic-lora-dubit-0.9.safetensors)

Compatibility: LTX-2.3.

For setup, usage, and prompting instructions, see the Dub-It guide.

---

LTX IC-LoRA catalog

The model card is the source of truth for an adapter’s required inputs, prompt wording, loader, and settings. A model card trained on LTX-2.3 is not, by itself, a guarantee of LTX-2.5 compatibility.

Available with LTX-2.5

Some LTX-2.5 workflows intentionally use an LTX-2.3-trained adapter. Follow the workflow’s exact model selection rather than substituting a similarly named checkpoint.

LTX-2.3-only adapters