AI video generation with LTX-2.3 22B — text-to-video, image-to-video clips for video production. Use when generating video clips, animating images, creating b-roll, animated backgrounds, or motion con
1---2name: ltx23description: AI video generation with LTX-2.3 22B — text-to-video, image-to-video clips for video production. Use when…4---56# .claude/skills/ltx2/SKILL.md
L
SKILL.mdAgent skill
Transport
File
Status
Listed
SKILL.md
Install
The skill file stays in the publisher's repository.
Paste the copied command in a terminal. It installs this SKILL.md into that agent.
shterminal
1npx skills add digitalsamba/claude-code-video-toolkit --skill ltx2 -a cursor -y
Paste the copied command in a terminal. It installs this SKILL.md into that agent.
AI video generation with LTX-2.3 22B — text-to-video, image-to-video clips for video production. Use when generating video clips, animating images, creating b-roll, animated backgrounds, or motion content. Triggers include video generation, animate image, b-roll, motion, video clip, text-to-video, image-to-video.
Download .claude/skills/ltx2/SKILL.md from the publisher, or copy the path and open the repository. PlugMCP lists the skill and does not host the file.
Where does ltx2 live?
The publisher keeps it at https://github.com/digitalsamba/claude-code-video-toolkit.
Which agents can use ltx2?
Claude, Cursor, Codex, and Grok Build can read a SKILL.md you add from the publisher's repository. Grok Bot installs it when you send the skill's GitHub folder link in a chat.
From the publisher
SKILL.md
name
ltx2
description
AI video generation with LTX-2.3 22B — text-to-video, image-to-video clips for video production. Use when generating video clips, animating images, creating b-roll, animated backgrounds, or motion content. Triggers include video generation, animate image, b-roll, motion, video clip, text-to-video, image-to-video.
LTX-2.3 Video Generation
Generate ~5 second video clips from text prompts or images using the LTX-2.3 22B DiT model.
Runs on Modal (A100-80GB). Requires MODAL_LTX2_ENDPOINT_URL in .env.
LTX-2.5 is a deploy-time opt-in on the same endpoint; see LTX-2.5 (opt-in).
Quick Reference
# Text-to-video
uv run tools/ltx2.py --prompt "A sunset over the ocean, golden light on waves, cinematic" --output sunset.mp4
# Image-to-video (animate a still image)
uv run tools/ltx2.py --prompt "Gentle camera drift, soft ambient motion" --input photo.jpg --output animated.mp4
# Custom resolution and duration
uv run tools/ltx2.py --prompt "..." --width 1024 --height 576 --num-frames 161 --output wide.mp4
# Fast mode (fewer steps, quicker)
uv run tools/ltx2.py --prompt "..." --quality fast --output quick.mp4
# Reproducible output
uv run tools/ltx2.py --prompt "..." --seed 42 --output reproducible.mp4
Parameters
Parameter
Default
Description
--prompt
(required)
Text description of the video
--input
-
Input image for image-to-video
--width
768
Video width (divisible by 64)
--height
512
Video height (divisible by 64)
--num-frames
121
Frame count, must satisfy (n-1) % 8 == 0
--fps
24
Frames per second
--quality
standard
standard (30 steps) or fast (15 steps)
--steps
30
Override inference steps directly
--seed
random
Seed for reproducibility
--output
auto
Output file path
--negative-prompt
sensible default
What to avoid
--lora
none
Style LoRA preset. Currently: crt-terminal.
Style LoRAs
Style LoRAs bias the output toward a specific visual aesthetic. They're baked into the Modal image and selected per-request; switching LoRAs rebuilds the server pipeline.
On an LTX2_VERSION=2.5 deployment: the presets below were trained on LTX-2.3. crt-terminal was checked on 2.5 (2026-10-12) and does not reproduce its CRT look there: it changes the output, but the result is a photoreal scene with green text. Use a 2.3 deployment for it. Keys that don't match are skipped without an error, so compare any other 2.3 LoRA with 2.3 before relying on it.
crt-terminal — CRT / pixel-art terminals
Base: LTX-2.3 22B, trained by @lovis93 (Apache 2.0).
# Trigger word is auto-prepended — write the prompt normally
uv run tools/ltx2.py --lora crt-terminal \
--prompt "a terminal typing out \"\\$ claude --continue\" character by character in glowing green pixel font, scanlines, phosphor glow, low choppy frame rate, hacker mood" \
--output crt_claude.mp4
What the preset changes:
Prepends crtanim, to the prompt (the LoRA's trigger word)
Defaults to 1024×1024, 121 frames (the ratio it was trained on)
Relaxes the default negative prompt so on-screen text isn't filtered out
Prompt pattern:<CRT aesthetic> → <color palette> → <animation style> → <subject> → <literal text in quotes> → <mood>. Keep on-screen text to 1–3 words — the model can't render long strings reliably. The LoRA prefers static framing; ask for camera moves explicitly if you want them.
Keep prompts under 200 words. Be specific about the scene.
Good Prompts
# Atmospheric b-roll
"Aerial drone shot slowly flying over turquoise ocean waves breaking on white sand, golden hour sunlight, cinematic"
# Product/tech scene
"Close-up of hands typing on a mechanical keyboard, shallow depth of field, soft desk lamp lighting, cozy atmosphere"
# Abstract background
"Dark moody abstract background with flowing blue light streaks, subtle geometric grid, bokeh particles floating, cinematic tech atmosphere"
# Animate a portrait
"Professional headshot, subtle natural head movement, confident warm expression, studio lighting, shallow depth of field"
# Animate a slide/screenshot
"Gentle subtle particle effects floating across a presentation slide, soft ambient light shifts, very slight camera drift"
Bad Prompts
# Too vague
"A cool video"
# Too many competing ideas
"A cat riding a skateboard while juggling fire on the moon during a thunderstorm"
# Describing text/UI (model can't render text reliably)
"A website showing the text 'Welcome to our platform'"
Video Production Use Cases
B-Roll Clips
Generate atmospheric 5s shots for cutaways between narrated scenes:
uv run tools/ltx2.py --prompt "Futuristic holographic interface, glowing data visualizations, clean workspace, cinematic" --output broll_tech.mp4
uv run tools/ltx2.py --prompt "Aerial view of European city at golden hour, modern architecture" --output broll_europe.mp4
Animated Slide Backgrounds
Feed a slide screenshot and add subtle motion:
uv run tools/ltx2.py --prompt "Gentle particle effects, soft ambient light shifts, very slight camera drift" --input slide.png --output animated_slide.mp4
Animated Portraits
Bring still headshots to life:
uv run tools/ltx2.py --prompt "Subtle natural head movement, warm expression, professional lighting" --input headshot.png --output animated_portrait.mp4
Keeping a real person the same person (learned animating colleagues' profile photos, 2026-11):
Feed the uncropped portrait (704x768); a tight 16:9 crop of a face makes LTX hard-cut to a stranger. Crop in
Remotion with objectFit: cover instead.
Lead with "Static camera, medium shot, framing stays exactly the same, no zoom, no cuts" (not "tripod": it
can draw one into the frame), end
with "the same person, the same face and clothes throughout, no text". Avoid "opening credits", "camera pushes
in" and "title".
One gentle action per clip. About one clip in three still needs a re-roll or a trim, so check every clip with
contact_sheet.py clip.mp4 --count 6. Full method: .claude/skills/parody-short/people.md.
Stylized Character Cameo (SadTalker Alternative)
For non-realistic faces — fantasy characters, masked figures, heavy beards, helmets, illustrations — SadTalker often produces uncanny or broken lip sync because it's trained on photoreal humans. LTX-2 image-to-video is frequently a better choice when lip-sync precision isn't critical (the viewer's brain fills in the gap as long as something is moving). Prompt for motion + atmosphere, not phonemes:
Heavy facial hair or accessories obscuring the mouth
Masked or helmeted figures
Short cameo lines where atmosphere matters more than precision
Dramatic VO rather than dialogue
When SadTalker still wins:
Photoreal human presenters
Full sentences where mouth shape needs to match phonemes
Tutorials / talking-head explainers where the viewer is effectively reading lips
Branded Intro/Outro
Generate abstract motion backgrounds for title cards:
uv run tools/ltx2.py --prompt "Dark moody background with flowing blue and coral light streaks, bokeh particles, cinematic tech atmosphere, no text" --output intro_bg.mp4
Combining with Other Tools
LTX-2 generates raw clips. Combine with the rest of the toolkit:
Workflow
Tools
Generate clip → upscale
ltx2.py → upscale.py
Generate clip → add to Remotion
ltx2.py → use as <OffthreadVideo> in composition
Generate image → animate
flux2.py → ltx2.py --input
Generate clip → extract audio
ltx2.py → ffmpeg -i clip.mp4 -vn audio.wav
Generate clip → add voiceover
ltx2.py → mix with qwen3_tts.py output
Technical Details
Model: LTX-2.3 22B DiT (Lightricks), bf16
GPU: A100-80GB on Modal (~$2.50/hr)
Inference: ~2.5 min per clip (768x512, 121 frames, 30 steps)
Cost: ~$0.07 per 5s clip warm, ~$0.12 cold at $2.50/hr (benchmark in docs/ltx2.md)
Weights: ~80GB baked into the image. Not kept in VRAM between requests: every request (warm or cold) loads the text encoder and the transformer for each stage, then frees them
Output: H.264 MP4 with synchronized ambient audio (24fps)
Max duration: ~8s (193 frames) per clip
Known Limitations
Training data artifacts: ~30% of generations may have unwanted logos/text from training data. Re-run with different --seed.
Text rendering: Cannot reliably generate readable text in video. Use Remotion overlays instead.
Max duration: ~8s per clip. Longer content needs stitching.
Audio: Generated audio is ambient/environmental only. Use voiceover/music tools for speech and music.
License: Community License — free under $10M revenue, commercial license needed above that.
Setup
# 1. Create Modal secret for HuggingFace (one-time)
uv run modal secret create huggingface-token HF_TOKEN=hf_your_token
# 2. Deploy (downloads ~80GB of weights, takes ~10 min)
uv run modal deploy docker/modal-ltx2/app.py
# 3. Save endpoint URL to .env
echo "MODAL_LTX2_ENDPOINT_URL=https://yourname--video-toolkit-ltx2-ltx2-generate.modal.run" >> .env
# 4. Test
uv run tools/ltx2.py --prompt "A candle flickering on a dark table, cinematic" --output test.mp4
Important: HuggingFace token needs read-access scope. Its account needs access to Google's gated Gemma 3 text encoder; Google approves these requests by hand, so ask before deploying. Unauthenticated downloads are severely rate-limited.
LTX-2.5 (opt-in)
LTX2_VERSION=2.5 at deploy time swaps LTX-2.3 for LTX-2.5 behind the same endpoint, request format and tool. LTX-2.3 stays the default: benchmarked against it, 2.5 is ~25-45% slower per clip and the crt-terminal LoRA only works on 2.3 (see docs/ltx2.md).
# Git Bash / macOS / Linux. In PowerShell: $env:LTX2_VERSION='2.5' first, and
# Remove-Item Env:LTX2_VERSION before deploying 2.3 again.
LTX2_VERSION=2.5 uv run modal deploy --strategy recreate docker/modal-ltx2/app.py
# Back to 2.3 (its cached image is reused, nothing rebuilds)
uv run modal deploy --strategy recreate docker/modal-ltx2/app.py
Access:Lightricks/LTX-2.5 is gated. Click "Agree and Access" on the model page with the token's account first (approved automatically; it shares your contact info). It bundles a Gemma 4 text encoder, so 2.5 doesn't need the Gemma 3 access.
License: LTX-2.x Community License. Free under $10M annual revenue, commercial use included; don't remove watermark or provenance features.
Check: the tool prints Model: Lightricks/LTX-2.5 (or Lightricks/LTX-2.3) after each clip.
LoRAs:crt-terminal doesn't keep its look on 2.5, see Style LoRAs above.
Speed: about 25-45% slower per clip than 2.3 (benchmark in docs/ltx2.md).
Details, differences and the benchmark plan: docs/ltx2.md.