Short answer: yes, LTX 2.5 is uncensored on our API, but check the anatomy. Based on our September 2026 tests, it returned 21 of 22 clips as asked from L1 (suggestive) up to L4 (an intimate scene between two adults; one came back toned down), and its L5 restraint clip (a consensual restraint scene between adults) came back whole. Several upper-level clips were rough: head cropped out, colour casts, merged limbs.
The L5 restraint probe came back with every element we named; only the setting changed, a stone cellar in amber light instead of a bed in red light. A 5-second 720p clip costs $0.45. The cheaper LTX 2.3 Spicy and Spicy LoRA delivered that probe in 2 of 4 and 3 of 4 runs, and softened the L5 violence probe every time.
The 30-second version
- LTX 2.5 delivered the L5 restraint probe. It is the only LTX here with text-to-video, 4K and sound.
- LTX 2.3 Spicy and Spicy LoRA are cheaper but uneven at the top. They only animate images, come back silent, and a softened clip is still billed.
- Pick LTX 2.5 when you start from words or need length, 4K or sound. Pick LTX 2.3 Spicy for the cheapest clip from a still, and plan on re-rolling the top rungs. Pick Spicy LoRA only for its built-in LoRA controls.
How far each LTX build goes
What the levels mean: L1 suggestive imagery, such as lingerie or a slipping strap · L2 moderate NSFW imagery · L3 advanced NSFW imagery · L4 an intimate scene between two adults · L5 the extreme tier, with two probes: a consensual restraint scene between adults, and a violent scene. ✅ came back as written · ⚠️ some runs toned down, with the count that came back as written · ❌ refused.
| Model · task | L1 suggestive | L2 moderate NSFW | L3 advanced NSFW | L4 two adults | L5 restraint | L5 violence | Page label |
|---|---|---|---|---|---|---|---|
| LTX 2.5 · text-to-video | ✅ 3 of 3 | ✅ 3 of 3 | ✅ 2 of 2, head cropped out | ⚠️ 2 of 3: one toned down | ✅ 1 of 1 | — | Uncensored |
| LTX 2.5 · image-to-video | ✅ 3 of 3 | ✅ 3 of 3 | ✅ 2 of 2 | ✅ 3 of 3 | ✅ 1 of 1 | — | Uncensored |
| LTX 2.3 Spicy · image-to-video | ✅ 3 of 3 | ✅ 3 of 3 | ✅ 3 of 3 | ⚠️ 2 of 3 | ⚠️ 2 of 4 | ⚠️ 0 of 3 | Uncensored |
| LTX 2.3 Spicy LoRA · image-to-video | ✅ 3 of 3 | ⚠️ 1 of 3 | ✅ 3 of 3 | ⚠️ 2 of 3 | ⚠️ 3 of 4 | ⚠️ 0 of 3 | Uncensored |
Nothing was refused. The one ⚠️ on LTX 2.5 is an L4 text-to-video clip whose action was only implied. The 2.3 builds were run from a still at 480p and 3 seconds; on the violence probe they toned it down on all three runs. Uncensored on our model pages means we add no content filter; the model itself still decides what comes back.
Softening is not linear: Spicy LoRA softened L2 in 2 of 3 runs but delivered L3 3 of 3 times. Run records: LTX 2.5, LTX 2.3 Spicy, LTX 2.3 Spicy LoRA.
What we saw in the clips
From an upper-level frame (LTX 2.5, L5). For all 5 seconds the frame stayed exactly at the level it went in. The catch: the clip barely moved, just a head tilt and a shift of gaze (fix below).
L4 on request (LTX 2.5). From an L3 still, all three clips moved on to the L4 action we wrote; from text, two of three did. The weak point was the picture: limbs merged, one clip opened on an abstract blob, red and orange casts washed over several, and both L3 clips from text cropped her head out of frame.
L2 on a stylised character (LTX 2.3 Spicy LoRA). A woman in a circuit-lined bodysuit, seen from behind, looks back at the camera and the 3-second clip moves up to L2. The neon sign from the still stays legible.
The everyday look.
Are 2K, 4K and sound worth paying for?
Sound: only LTX 2.5 has it, and you can't switch it off. The level varies, from clear to faint. Both 2.3 builds return a track that is silent or close to it. If the clip goes out with sound, LTX 2.5 saves you the foley; if you score it yourself, the 2.3 builds lose nothing.
2K: probably, for the take you keep. LTX 2.5 at 2k renders 2560 × 1408, with crisp skin texture and loose strands of hair at full size.
4K: only when the deliverable needs it. 5 seconds costs $0.45 at 720p, $0.95 at 2k and $1.50 at 4K. Draft at 720p until the motion is right, then re-run the same prompt and seed at 2k or 4K. The 2.3 builds stop at 1080p.
A prompt can ask for sound directly, here an orchestra tuning up:
How to write LTX prompts
Write the shot, not the vibe: who is in frame and what they're wearing, what they do in order, what the camera does, the light, and on LTX 2.5 the sound. Plain words work; no euphemisms or "artistic" framing needed.
Lightricks' prompting guide agrees: one present-tense paragraph of four to eight sentences, active verbs, one light source, an uncrowded frame, no on-screen text. Adapt the prompts below to your own scene.
A 20-second solo scene (LTX 2.5 text-to-video)
One continuous 20-second take, no cuts. An adult woman in a black silk robe stands at a rain-streaked hotel window at night, her back to the camera. She pulls the robe closer around her shoulders, then turns from the glass and walks slowly toward the camera, stopping close to hold the lens with a calm look. The camera pushes in slowly, then holds. Warm lamplight only, 35mm grain. Rain on the glass and a low city hum.
Twenty seconds is long: give it three or four beats joined by "then", in the order they happen.
Two people, close (LTX 2.5 text-to-video)
Two adults, a man and a woman, lie facing each other under a loose white sheet in a sunlit loft at dawn. She traces a slow line down his arm; he pulls her closer and she rests her forehead against his. The camera circles the bed once at mattress height. Soft window light, dust in the air, 35mm grain. Birdsong outside, quiet breathing, no music.
Say "two adults" and who is who. One shared action and one camera move per clip.
Animating your own still (LTX 2.5 image-to-video)
The opening frame sets the people, the wardrobe, the set and the shape of the clip. The prompt, which LTX 2.5 image-to-video requires, has one job: movement. Write actions, not a description of the picture.
She pushes up from the pillows, turns toward the camera and gathers the sheet around her shoulders, then leans back on her hands, tips her head and holds the look. Slow push-in, then hold. The lamp flickers once. Rain on the window, quiet room tone.
Run this prompt on LTX 2.5 image-to-video →
Start and end frames (LTX 2.5 image-to-video)
LTX 2.5 takes a second still as the closing frame. It lands on that frame, but a big change between two very different stills plays as a dissolve or a snap rather than a move. Example: a couple on a beach at sunset and the same couple by a night bonfire.
In 10 seconds with a prompt about the sun going down, the beach holds for two seconds, then dissolves into the bonfire like a cross-fade in an edit. With 5 seconds and a prompt about the surf, the clip plays the surf and jumps to the bonfire in its last quarter-second:
For one continuous move, keep both frames in the same room, the same light and the same outfit, change only the pose, and describe the path between them:
Same bedroom, same lamp light, same woman in both frames. She walks in from the doorway, sits on the edge of the bed and lies back into the pose of the final frame. One continuous move, no cuts, the camera tracking beside her. Bare feet on a wooden floor, rain outside.
Run this prompt on LTX 2.5 image-to-video →
Short prompt, or none (LTX 2.3 Spicy)
LTX 2.3 Spicy takes a still with no prompt at all and adds a small movement. When you want a specific beat, keep it this short:
Slow breath. She arches her back and stretches both arms above her head, then turns her face to the lens and holds it. Locked-off camera.
Run this prompt on LTX 2.3 Spicy →
Both 2.3 builds also have a preset field: tuned is the default, and original styles less (on LTX 2.3 Spicy, almost not at all), so switch to it when the preset fights the look you wrote.
Built-in LoRAs (LTX 2.3 Spicy LoRA)
Spicy LoRA is LTX 2.3 Spicy plus 13 built-in LoRAs. Add up to three in a loras list, each a path (the LoRA's key) and a scale from 0 to 4 (0 turns it off, 1 is the default), on top of the preset. Your own LoRA files aren't accepted. The default mix softened L2 more often than plain Spicy. One look LoRA at scale 1:
"loras": [{ "path": "synth", "scale": 1 }]better_motion, physics_v2, hardcut and transition change how the clip moves rather than how it looks. Use one look LoRA near 1 with one motion LoRA lower, for example [{ "path": "synth", "scale": 1 }, { "path": "better_motion", "scale": 0.5 }], and change one LoRA at a time with frame, prompt and seed fixed.
Rain hisses on the rooftop. She finishes her turn, the glowing lines along her arms and spine pulsing in sequence, and looks back over her shoulder at the camera. Slow arc around her, neon fog rolling behind.
Run this prompt on LTX 2.3 Spicy LoRA →
Preparing the opening frame
- Make it with an uncensored image model and approve it before you animate. On image-to-video the frame decides who is on screen and what they wear.
- The frame sets the shape. A 9:16 still gives a 9:16 clip; image-to-video has no separate aspect setting.
- Crop out signs, screens and captions you don't want on screen; lettering in the still carries into the clip.
- For a closing frame, reuse the set and the light. The further apart the two stills, the more the clip dissolves instead of moving.
Where LTX trips up, and the fix
- Image-to-video that barely moves. The frame sets the look; the prompt must set the motion. Write two or three actions in order plus one camera move.
- Rough anatomy. LTX 2.5 can merge bodies, crop a head out and tint clips red or orange; users also report faces drifting across image-to-video clips. Name the framing you want ("full body, head to feet"), draft at 720p and re-roll before a long or 4K render; compare first and last frames when the same person must hold.
- The 2.3 builds are unpredictable at the top. A softened clip is billed. Draft 3 seconds at 480p for $0.057 on 2.3 Spicy, check it, and only then pay for 20. To stay close to what you wrote, use LTX 2.5.
- Text-to-video moves your set. Our bedroom became a stone cellar. When the set matters, put it in an opening frame and use image-to-video.
- LTX 2.5 starts at 5 seconds, and nothing goes past 20. Fix: take the last frame of a clip you like and use it as the next opening frame. The longer uncensored video guide walks through the chain.
- Square or 4:3 from text. LTX 2.5 text-to-video does 16:9 and 9:16 only. Fix: make a square still and use image-to-video.
Which LTX to pick
| Model | Reach | Starts from | Length | Top resolution | Sound | 5 s from | Pick it for |
|---|---|---|---|---|---|---|---|
| LTX 2.5 | L1 suggestive to L4 two adults: 21 of 22 clips; L5 restraint: 1 of 1 per task | Text, or an image plus an optional end frame | 5–20 s | 4K | Yes, always on | $0.45 (720p) | Text-to-video, long takes, 4K, sound |
| LTX 2.3 Spicy | L1 suggestive to L3 advanced clean; L4 two adults 2 of 3; L5 restraint 2 of 4 | Image, prompt optional | 3–20 s | 1080p | Silent | $0.095 (480p) | The cheapest clip from a still, with re-rolls at the top |
| LTX 2.3 Spicy LoRA | L1 suggestive and L3 advanced clean; L2 moderate 1 of 3; L4 two adults 2 of 3; L5 restraint 3 of 4 | Image, prompt optional | 3–20 s | 1080p | Near-silent | $0.1425 (480p) | Built-in LoRA controls, at 17–50% more per second than plain Spicy |
Which one? Words only, or you need sound, 4K or a pinned end frame: LTX 2.5. A folder of stills and a budget, where a softened draft costs a few cents: LTX 2.3 Spicy. At the top rungs plan on about two 3-second 480p drafts per keeper: roughly $0.11 on 2.3 Spicy, $0.17 on Spicy LoRA, before the final render. Spicy LoRA only for its built-in LoRA controls. If the L5 restraint scene is the point of the job, LTX 2.5 is the better bet; if one face must stay recognisable through the clip, the 2.3 builds are the safer pick.
Against MiniMax H3: H3 is cheaper ($0.185 for 5 seconds at 768p) and reaches L5 straight from text; LTX 2.5 goes longer (20 seconds against 15) and higher (4K against 2k). See our H3 test and LTX 2.5 vs MiniMax H3.
If you meant the open weights, or your own LoRA
Search results for "LTX 2.5 NSFW" are mostly local builds: the weights are open under Lightricks' LTX-2.x community licence, which its LTX 2.5 page describes as free below $10M annual revenue, and community uncensored text encoders and LoRAs for ComfyUI circulate on Hugging Face and Civitai. Local users praise speed and the licence, rate MiniMax H3 above LTX 2.5 on quality, and disagree on whether LTX 2.3 LoRAs carry over. The hosted route is a request instead of a GPU and a workflow: 20 seconds up to 4K with sound, priced per second. It doesn't accept your own LoRA files: LTX 2.5 has no LoRA field, and Spicy LoRA has 13 built-in LoRAs you can only re-weight.
What a clip costs
| 5 seconds | 20 seconds | |
|---|---|---|
| LTX 2.5, 720p | $0.45 | $1.80 |
| LTX 2.5, 1080p | $0.65 | $2.60 |
| LTX 2.5, 2k | $0.95 | $3.80 |
| LTX 2.5, 4K | $1.50 | $6.00 |
| LTX 2.3 Spicy, 480p | $0.095 | $0.38 |
| LTX 2.3 Spicy, 720p | $0.19 | $0.76 |
| LTX 2.3 Spicy, 1080p | $0.285 | $1.14 |
| LTX 2.3 Spicy LoRA, 480p | $0.1425 | $0.57 |
| LTX 2.3 Spicy LoRA, 720p | $0.2375 | $0.95 |
| LTX 2.3 Spicy LoRA, 1080p | $0.3325 | $1.33 |
The cheapest draft is 3 seconds of LTX 2.3 Spicy at 480p: $0.057. LoRA settings don't change the price. At 720p LTX 2.3 Spicy costs 42% of LTX 2.5 per second and Spicy LoRA 53%. A clip that finishes toned down is billed like any other. Failed tasks are normally free, and a refusal is charged only when the model page says so; none of the three LTX pages does (see the billing policy). Median generation time is 85 seconds for LTX 2.5, 66 for LTX 2.3 Spicy and 63 for Spicy LoRA. Each model page shows the live estimate before you submit.
Try it in the Playground, then the API
In the Playground. Create an account, then:
- Open LTX 2.5 text-to-video (or image-to-video, LTX 2.3 Spicy, Spicy LoRA) and use the Playground tab.
- Paste a prompt from above into Prompt. On image-to-video, add your opening frame under Input image; LTX 2.5 also takes an optional Closing frame.
- Set Resolution and Duration, both marked "affects price". Draft at 720p and 5 seconds on LTX 2.5, or 480p and 3 seconds on the 2.3 builds.
- Check the estimate shown for your settings and press Run model. Reuse the Seed when you step up to 2k or 4K. To work in a browser workspace instead of a model page, Spicy Studio makes images and short videos with no code.
Then the API, once a prompt works and you want it in your own app. This is LTX 2.5 text-to-video; image-to-video (any build) takes an image_url too.
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $(uuidgen)" \
--data '{
"model": "lightricks/ltx-2.5/text-to-video",
"input": {
"prompt": "One continuous take. An adult woman in a black silk robe stands at a rain-streaked hotel window at night, then turns toward the camera. Slow push-in, then hold. Warm lamplight only. Rain on the glass and a low city hum.",
"resolution": "720p",
"duration_seconds": 5
}
}'Fields, results and errors are in the API reference. Results expire, so see retention and delivery before you build on the returned link.
Or let an Agent run it. An agent can make the same call through the CLI, the MCP server or the Agent Skill. The CLI and MCP server read the live model schema, show the price and ask you to confirm before a paid task. With SPICY_API_KEY set:
# CLI
npx @spicyapi/cli tasks create --model lightricks/ltx-2.5/text-to-video \
--input-json '{"prompt": "An adult woman in a black silk robe stands at a rain-streaked hotel window at night, then turns toward the camera. Slow push-in. Rain on the glass.", "resolution": "720p", "duration_seconds": 5}' --wait
# MCP server: the command your assistant's MCP config runs; then ask it:
# "Run LTX 2.5 text-to-video with this prompt, 720p, 5 seconds."
# It calls spicyapi_task_create, which stops for your price confirmation, then spicyapi_task_wait.
npx --yes --package=@spicyapi/mcp spicyapi-mcp
# Agent Skill
npx skills add spicyapi-ai/spicy-skillFAQ
Is LTX 2.5 uncensored?
On our API, yes, on both text-to-video and image-to-video. We add no content filter; the L5 restraint prompt came back as written, and 21 of 22 further clips came back as asked up to L4.
Does LTX 2.5 go to L4 when you ask?
Yes: 5 of 6 clips delivered it (3 of 3 from a still, 2 of 3 from text), with rough anatomy in several. LTX 2.3 Spicy and Spicy LoRA delivered L4 in 2 of 3 runs each.
Does LTX make sound? Can I turn it off?
LTX 2.5 always generates a soundtrack; there's no switch, so describe the sound you want or mute it in the edit. The 2.3 builds come back silent or close to it.
How do I use a start and an end frame?
On LTX 2.5 image-to-video, add the closing still next to the opening one. Keep both in the same set, light and outfit, and describe the path between them; very different stills give a dissolve, not a move.
Is LTX 2.3 Spicy just a cheaper LTX 2.5?
No. It's an uncensored, LoRA-tuned build of LTX 2.3: image-to-video only, up to 1080p, no sound, at 42% of LTX 2.5's price per second at 720p. Its top rungs are uneven (L4 2 of 3 runs; L5 restraint 2 of 4).
My LTX image-to-video comes out almost still. Why?
The prompt probably describes the picture instead of the movement. Write two or three actions in order and a camera move.
Do I pay for a refused or softened clip?
A softened clip is billed like any other. A refused or failed task is normally free unless the model page says otherwise; none of the three LTX pages does. Check the clip, not just the status.
Can I use my own LoRA with LTX 2.5, or run it locally?
Not through our API: LTX 2.5 has no LoRA field, and Spicy LoRA accepts only its 13 built-in LoRAs. The open weights can be run locally.
Can I use a photo of a real person?
Only with that person's consent, and only for the use they agreed to.
Your part
We don't put a content filter in front of these models, so nobody checks your inputs for you. Everyone on screen must be an adult. Any real, identifiable person must have consented. You need the rights to everything you upload, and you're responsible for the law where you publish. If you self-host the open LTX weights instead, read Lightricks' LTX-2 Community License and Acceptable Use Policy yourself; the policy has a section headed "Do Not Generate Sexually Explicit Content", and whether and how it applies to your setup is yours to check.
Run your first clip
Open LTX 2.3 Spicy in the Playground with a still of your own and run a 3-second 480p draft for $0.057. When you need text-to-video, sound or 4K, paste the 20-second solo prompt into LTX 2.5 text-to-video.
