LTX 2.5 · self-hosted

LTX 2.5 video model, self-hosted on Wavemaker's GPU fleet. Pick it in the model picker or name it in your brief. This page gathers its model configurations and example graphs.

Discover · Wavemaker catalog

When to use it

  • LTX 2.5 video model, self-hosted on Wavemaker's GPU fleet. Pick it in the model picker or name it in your brief. This page gathers its model configurations and example graphs.

What you provide

  • One owned video clip, the seconds to add (2-20), start or end, and a prompt describing the whole shot including the new seconds.
  • One owned video clip, the seconds to add, start or end, and a prompt.
  • One owned video clip (up to 257 frames per pass at 1080p or below). A prompt describing the empty result and naming what must go.
  • One owned video clip (9 frames or more; up to 257 frames per pass at 1080p or below). A prompt describing the scene and the colours wanted.

What you get

  • The clip plus the new seconds (total snapped to 8n+1 frames), same width, height and frame rate, with sound.
  • The clip plus the new seconds, same geometry and frame rate.
  • The same clip without the transient subjects: identical width, height, frame count and frame rate, with the source audio.
  • The same clip in colour: identical width, height, frame count and frame rate, with the source audio.

Creative guidance

  • LTX-2.3 Extend: Adds 2-20 seconds after the end or before the start of your clip, continuing its scene, motion and sound.
  • LTX-2.3 Extend (HQ): The extension on the full LTX-2.3 dev model with 30 steps.
  • LTX-2.3 Clean Plate: Removes passers-by, traffic and clutter from a shot and keeps the location, light, camera and sound.
  • LTX-2.3 Colorization: Turns black-and-white or faded footage into natural colour at the same size, length, framing and soundtrack.
  • LTX-2.3 Cross-Eyed: A novelty crossed-eye look on the same person in the same clip, for comedy and meme creative.
  • LTX-2.3 Day to Night: The same shot at night: dark sky, streetlights on, with the layout, people and motion kept.
  • LTX-2.3 Decompression: Cleans macroblocks, banding and chroma bleed out of a re-encoded or low-bitrate clip at the same size, length, framing and soundtrack.
  • LTX-2.3 Detail Finish: Restores crisp fine detail to a soft clip at the same size, length, framing and soundtrack, as a finish before an upscale.
  • LTX-2.3 Dub-It: A new line in the same shot: new words, new lip movements and the speaker's own voice, with the face and framing kept.
  • LTX-2.3 Talking Greenscreen: Puts a talking presenter on flat chroma-key green with the same performance and sound, ready to key onto any set.
  • LTX-2.3 Inpaint (boxes): Replaces or removes what is inside one or two boxes of a clip and keeps everything else, pixel for pixel, with the same sound.
  • LTX-2.3 Inpaint (mask video): Regenerates the white area of a mask video in a clip and keeps everything else, pixel for pixel, with the same sound.
  • LTX-2.3 Instant Shave: The same person, clean-shaven, in the same clip: identity, expression, framing and sound unchanged.
  • LTX-2.3 Expand / Outpaint: Turns a 16:9 master into 9:16, 1:1 or a wider frame by generating new scene around the untouched source; different from cropping.
  • LTX-2.3 Pixel Upscale x2: Generative 2x upscale of a small AI draft: re-renders the clip with crisp texture and keeps the framing, length and sound.
  • LTX-2.3 Pixel Upscale x4: Generative 4x upscale of a small AI draft: re-renders the clip with crisp texture and keeps the framing, length and sound.
  • LTX-2.3 Water Simulation: Adds a water surface to a scene with the same camera, motion and sound; strongest as standing water on flat ground.
  • LTX-2.3 Retake: Fixes a few seconds of a good clip in place: regenerates the window you name, keeps every other frame and the soundtrack.
  • LTX-2.3 Retake (HQ): The retake on the full LTX-2.3 dev model with 30 steps: for when the fast retake shows artefacts.
  • LTX 2.3 · audio-conditioned replacement: Use the explicitly selected LTX 2.3 audio-conditioned graph that replaces the unwired LTX 2.0 source; the replacement identity remains visible.
  • LTX 2.0 · edge-guided restyle: Restyle your own clip while its outlines and motion stay put: LTX-2.0 with the canny IC-LoRA, steered by edges of the source and a restyled first frame.
  • LTX 2.0 · depth-guided restyle: Restyle your own clip while its geometry and camera move stay put: LTX-2.0 with the depth IC-LoRA, steered by a depth map of the source and a restyled first frame.
  • LTX 2.0 · pose-guided restyle: Give a clip's choreography to a new character: LTX-2.0 with the pose IC-LoRA, steered by SDPose skeletons of the source and a restyled first frame.
  • LTX-2.3 i2v · style and motion LoRAs: LTX-2.3 image to video with two LoRA slots for a cinemagraph, an FPV drone glide or firmer physics, with sound.
  • LTX 2.3 · image and audio conditioning: Combine an owned image and audio reference in the explicit native LTX 2.3 image/audio-to-video topology.
  • LTX 2.3 · union-control restyle: Restyle your own clip with LTX-2.3 and the union-control IC-LoRA, steered by a MoGe depth map of the source and a restyled first frame; one fast pass.
  • LTX 2.3 · talking clip from a face and a voice: A talking clip of the person in an image, speaking a line you write in the voice of a reference recording: LTX-2.3 with the ID-LoRA.
  • LTX 2.5 video · model service: Use a configured LTX 2.5 service for text- or image-conditioned video after choosing the matching task example.
  • LTX 2.5 image animation · native: Animate an owned image with the distilled LTX 2.5 two-stage graph; text-only and first/last-frame tasks have separate profiles.
  • ltx25 native 4k i2v: ltx model configuration for image-to-video.
  • ltx25 native a2v: ltx model configuration for a2v.
  • LTX 2.5 · native first/last-frame video: Connect owned start and end images with the task-specific distilled LTX 2.5 topology.
  • ltx25 native keyframes: ltx model configuration for first/last-frame video.
  • ltx25 native long: ltx model configuration for image-to-video.
  • ltx25 native multishot: ltx model configuration for image-to-video.
  • LTX 2.5 · native text-to-video: Use the task-specific distilled LTX 2.5 text-video topology and its explicit batch, resize and guide defaults.
  • LTX-2.3 Extend: Adds 2-20 seconds before or after your clip with LTX-2.3, continuing its scene, motion and sound across a feathered half-second seam.
  • LTX-2.3 Extend (HQ): The LTX-2.3 extension on the 22B dev checkpoint with 30 steps and full guidance.
  • LTX-2.3 Clean Plate (IC-LoRA): Removes passers-by and transient clutter from a shot while keeping the location, light and camera: LTX-2.3 regenerates the same frames with the Lightricks Clean Plate IC-LoRA, using your clip as the guide, and keeps…
  • LTX-2.3 Colorization (IC-LoRA): Turns black-and-white or faded footage into natural colour: LTX-2.3 regenerates the same frames with the Lightricks Colorization IC-LoRA, using your clip as the guide, and keeps its size, length, frame rate and…
  • LTX-2.3 Cross-Eyed (IC-LoRA): Gives the person in a clip a novelty crossed-eye look with the Lightricks Cross-Eyed IC-LoRA on LTX-2.3, keeping identity, speech, size, length and soundtrack.
  • LTX-2.3 Day to Night (IC-LoRA): Turns a daylight shot into the same shot at night with the Lightricks Day-to-Night IC-LoRA on LTX-2.3, using your clip as the guide and keeping its layout, motion, size, length and soundtrack.
  • LTX-2.3 Decompression (IC-LoRA): Cleans macroblocks, banding and chroma bleed out of a re-encoded or low-bitrate clip without changing it: LTX-2.3 regenerates the same frames with the Lightricks Decompression IC-LoRA, using your clip as the guide, and…
  • LTX-2.3 Detail Finish (Deblur IC-LoRA): Sharpens a soft or slightly blurred clip without changing it: LTX-2.3 regenerates the same frames with the Lightricks Deblur IC-LoRA, using your clip as the guide, and keeps its size, length, frame rate and soundtrack.
  • LTX-2.3 Dub-It (IC-LoRA): Changes what a presenter says in an existing talking-head shot: LTX-2.3 with the Lightricks Dub-It IC-LoRA regenerates the mouth and the voice for a new line in one pass, keeping the face, framing, motion and the…
  • LTX-2.3 Expand / Outpaint (In-Outpainting IC-LoRA): Turns a clip into a larger or differently shaped frame (16:9 to 9:16, 1:1 or wider) by generating new scene around the untouched source instead of cropping it, keeping its length, frame rate and soundtrack (Lightricks…
  • LTX-2.3 Talking Greenscreen (IC-LoRA): Re-renders a talking presenter on a flat chroma-key green with the same performance, lip motion and soundtrack, ready to key onto any background (community greenscreen-avatar IC-LoRA on LTX-2.3).
  • LTX-2.3 Inpaint by Boxes (In-Outpainting IC-LoRA): Regenerates one or two boxes of a clip, for a time window or the whole clip, and keeps every other pixel and the soundtrack: LTX-2.3 with the Lightricks In-Outpainting IC-LoRA in two stages, then the source composited…
  • LTX-2.3 Inpaint by Mask Video (In-Outpainting IC-LoRA): Regenerates the white area of a mask video (any shape, moving or still) and keeps every other pixel and the soundtrack: LTX-2.3 with the Lightricks In-Outpainting IC-LoRA in two stages, then the source composited back.
  • LTX-2.3 Instant Shave (IC-LoRA): Shows the same person clean-shaven in the same clip: LTX-2.3 regenerates the frames with the Lightricks Instant Shave IC-LoRA, using your clip as the guide, and keeps identity, expression, size, length and soundtrack.
  • LTX-2.3 Pixel Spatial Upscale x2 (IC-LoRA): Upscales a clip 2x by re-rendering it with LTX-2.3 and the Lightricks Pixel Spatial Upscaler IC-LoRA, using your clip as the guide: crisp new texture, same framing, length, frame rate and soundtrack. Generative, so it…
  • LTX-2.3 Pixel Spatial Upscale x4 (IC-LoRA): Upscales a clip 4x by re-rendering it with LTX-2.3 and the Lightricks Pixel Spatial Upscaler IC-LoRA, using your clip as the guide: crisp new texture, same framing, length, frame rate and soundtrack. Generative, so it…
  • LTX-2.3 Water Simulation (IC-LoRA): Adds a water surface to a scene with the Lightricks Water Simulation IC-LoRA on LTX-2.3, using your clip as the guide and keeping its size, length, frame rate and soundtrack; strongest as standing water on flat ground.
  • LTX-2.3 Retake: Regenerates a time window of your own clip (video, sound or both) with LTX-2.3 and keeps every frame outside it, at the same size, length and frame rate.
  • LTX-2.3 Retake (HQ): The LTX-2.3 retake on the 22B dev checkpoint with 30 steps and full guidance: slower, for when the fast retake shows artefacts.
  • LTX-2.3 image to video with style LoRAs (cinemagraph, FPV, physics): Animates a still with LTX-2.3 (two-stage distilled, with sound) and up to two style or motion LoRAs: a cinemagraph where one element moves, a smooth FPV drone glide, or firmer object physics.
  • LTX-2.5 4K from a still: A Comfy graph that animates a 4K still into a 3840x2160 LTX-2.5 take (two stages, x2 latent upscale): packshots and other short 4K moves.
  • LTX-2.5 audio to video: A Comfy graph that animates a still to a supplied voice, song or ambience with LTX-2.5: the audio is held fixed and carried unchanged, and only the picture is generated.
  • LTX-2.5 keyframes at chosen frames: A Comfy graph that moves one continuous LTX-2.5 take through three keyframes placed on chosen frames: motion between approved storyboard frames, with sound.
  • LTX-2.5 looping 30 s take: A Comfy graph that makes a silent 30 s LTX-2.5 take from a still with ComfyUI-LTXVideo's looping sampler, one prompt per temporal tile; it waits for an LTX-2.5 variant with the looping sampler.
  • LTX-2.5 multishot from a keyframe: A Comfy graph that makes one LTX-2.5 take of 2-4 shots written in prose, opening on a reviewed still: the official two-stage distilled settings with sound.
  • Hosted LTX-2.5 image-to-video (engine): Hosted LTX-2.5 image-to-video (engine). Authored Denoise Engine node; standard rail.
  • LTX best_seed + LoRA pins: LTX best-seed with LoRA pins — the best-seed race with style/subject LoRA adapters loaded, for runs that need a specific look locked in.
  • LTX IC canny: An LTX video graph using detected image edges to guide the structure of generated frames.
  • LTX IC depth: An LTX video graph using a depth representation to guide scene geometry during generation.
  • LTX IC detailer: An LTX finishing graph that applies a detail and deblur treatment to a supplied video.
  • LTX IC inpaint: An LTX editing graph for repairing or changing a masked region of a supplied video.
  • LTX IC outpaint: An LTX editing graph for extending the visible scene beyond a video's original frame.
  • LTX IC pose: An LTX 2.0 pose IC graph: SDPose skeletons of your clip guide the new take, and the skeleton video is saved too.
  • Identity LoRA: Identity-preserving clip. Requires likenessConsent=true.
  • Video inpaint: Masked spatial repair on a short clip (hosted compute inpaint bridge).
  • Seed hunter: Preview several cheap seeds, then finalize the winner. Pick is required before full spend.
  • LTX union control: LTX Union-Control (pose/depth/canny) capability leaf — /retake control path
  • LTX-2.0 native A2V: A Comfy graph for LTX 2.0 video generation conditioned on an audio track.
  • LTX-2.0 native canny: A Comfy graph for LTX 2.0 video generation guided by edge maps from reference material.
  • LTX-2.0 native depth: A Comfy graph for LTX 2.0 video generation guided by depth information from reference material.
  • LTX-2.0 native pose: A Comfy graph for LTX 2.0 video guided by SDPose skeletons of your clip; it also saves the skeleton video.
  • LTX-2.3 native IA2V: Native Comfy image+audio-to-video. No hosted compute audio_url.
  • LTX-2.3 native IC: Native IC-LoRA control. Rejects novelty LoRAs.
  • LTX-2.3 native ID-LoRA: Identity LoRA native Comfy. likenessConsent required.
  • LTX 2.5 first-last-frame: LTX-2.5 first-last-frame style Comfy port (T-08) via single seed + start/end motion prompt
  • LTX 2.5 image-to-video: LTX 2.5 image-to-video: a seed still, then a short motion clip.
  • LTX-2.5 native 4K: An LTX 2.5 Comfy graph configured for a 4K output path, with substantial GPU and model requirements.
  • LTX-2.5 native first-last-frame video: Official Comfy first-last-frame LTX-2.5 on L40S.
  • LTX-2.5 native image-to-video: Official Comfy LTX-2.5 image-to-video on L40S.
  • LTX-2.5 native image-to-video 1080p: Official Comfy LTX-2.5 I2V at 1920×1080, 6 / 8 / 10 s (duration is the promoted PrimitiveInt 362).
  • LTX-2.5 native image-to-video 1440p: Official Comfy LTX-2.5 I2V at 2560×1440, 6 / 8 / 10 s (duration is the promoted PrimitiveInt 362).
  • LTX-2.5 native multishot: Native multishot text-to-video. Long-take stays video-ltx-longtake.
  • LTX-2.5 native text-to-video: A native Comfy graph for generating video from a text prompt with LTX 2.5.
  • LTX-2.5 native text-to-video 1080p: Official Comfy LTX-2.5 T2V at 1920×1080, 6 / 8 / 10 s (duration is the promoted PrimitiveInt 362).
  • LTX-2.5 native text-to-video 1440p: Official Comfy LTX-2.5 T2V at 2560×1440, 6 / 8 / 10 s (duration is the promoted PrimitiveInt 362).
  • LTX 2.5 text-to-video: LTX 2.5 text-to-video from a prompt.
  • Amateur phone still (Krea 2): Photoreal amateur-phone still on Krea 2 — casual capture energy, not a copied community workflow.
  • image-to-video animate: image-to-video animate — the default image-to-video path: bring one still to life with natural motion. The most-used single step on the platform.
  • Upscale+interpolate: Upscale + interpolate an existing clip (load video → LTX SR → 48fps). Does not generate a new clip.

Setup and requirements

  • Name the option you want in your brief; the agent reads its method or configuration (Source entry points) and applies it.

Source licenses and model terms continue to apply. Check runtime availability before generation.

Download source · Learn how to reuse resources

Related resources