LTX 2.3 Spicy Image to Video LoRA

Animate any image with a custom LoRA stack — slap, choke, physics, motion and more. Same cinematic spicy backbone as the base LTX 2.3 Spicy with fine-grained style and motion control.

📄 About LTX 2.3 Spicy Image to Video LoRA

LTX 2.3 Spicy Image to Video LoRA is the LoRA-controlled variant of the LTX 2.3 Spicy image-to-video backbone. It takes the same long-form motion quality — up to 20 seconds at 1080p — and exposes the underlying LoRA stack as a JSON object you can dial. Users often land on this page after searching for an ai image to video LoRA uncensored mode or an ai image to video generator LoRA uncensored toggle; to be clear up front, JAI Portal does NOT enable uncensored or unrestricted explicit output on this LoRA variant or anywhere else on the platform. What the LoRA controls actually unlock is fine-grained MOTION control: slap, choke, physics, better-motion, hardcut transition, dreamlike synth and any of the named LoRAs the upstream model ships with. You set each LoRA's strength between 0 and 1, and the model blends them on top of your chosen preset. Workflow: upload a JPG or PNG reference image, write the motion / camera prompt, optionally fill the LoRA strength object (e.g. <code>{"omninft": 0.5, "physics_v2": 0.8}</code>), pick resolution and duration, generate. If you leave the LoRA field blank, the model runs with the chosen preset's defaults — same behavior as the base <a href="/model/ltx-2-3-spicy-image-to-video">LTX 2.3 Spicy Image to Video</a> but with the LoRA pricing tier. Available LoRA keys include sulphur, sulphur_v1, vbvr, dreamly, synth, plora, singularity, omninft, omninft_bf16, better_motion, physics_v2, hardcut and transition. Unknown keys are ignored; setting a key to 0 disables that LoRA within the preset. Pricing follows the same per-second + resolution-multiplier shape as the base model but with a slightly higher starting rate (the LoRA stack adds GPU cost). 480p starts at the LoRA base, 720p is about 1.67×, 1080p is about 2.33×. The 5-second billing minimum is preserved so short iteration runs are bounded. Like the base model, you can browse the model page with no login — only generation consumes credits, and once logged in your session is cached for no login mid-workflow. Use this variant when the base LTX 2.3 Spicy keeps producing the same flavor of motion and you need to push it elsewhere. For example: dial physics_v2 up for grounded mass and momentum, dial better_motion up for smoother limb interpolation, dial omninft up for more stylized rendering. For pre-baked action LoRAs as one-tap workflows, the <a href="/model/jai-slap-lora">JAI Slap LoRA</a>, <a href="/model/jai-choke-lora">JAI Choke LoRA</a> and <a href="/model/jai-pose-lora">JAI Pose LoRA</a> models are dedicated single-motion workflows on the WAN backbone — useful when you want a known action with zero tuning, while LTX 2.3 Spicy LoRA is the option for open-ended motion with fine-grained control. For sibling spicy options on different backbones: <a href="/model/wan-2-2-spicy-image-to-video">WAN 2.2 Spicy Image to Video</a> is the best fast-iteration short-clip option, <a href="/model/vidu-q3-image-to-video-spicy">Vidu Q3 Image to Video Spicy</a> is the cheapest cinematic short clip generator, and <a href="/model/bytedance-seedance-v1-5-pro-image-to-video-spicy">Bytedance Seedance v1.5 Pro Image to Video Spicy</a> is the option to reach for when you need native audio alongside the video. LTX 2.3 Spicy LoRA is the most flexible — longest duration, most motion control — at the cost of being slightly slower per generation and requiring you to think about which LoRAs to dial. All paid generations include full commercial usage rights with no royalties, no attribution and no usage restrictions on the delivered output. Output is MP4, audio-free, and downloadable as soon as the queue completes.

✨ Key Features

Full LoRA stack control — set per-LoRA strength as a JSON object, mix multiple named LoRAs in a single generation

Same 20-second 1080p ceiling as the base LTX 2.3 Spicy — best duration × motion-quality combination in the spicy catalog

Preset toggle (tuned / original) gives you a baseline stack you can then override per-LoRA

Available LoRA keys: sulphur, vbvr, dreamly, synth, plora, singularity, omninft, omninft_bf16, better_motion, physics_v2, hardcut, transition — covers motion quality, physics, style and transitions

Unknown LoRA keys are ignored gracefully so you can experiment without breaking the request

Setting a LoRA to 0 disables it inside the preset — fine-grained subtractive control, not just additive

Per-second linear pricing with 5-second billing minimum so short iteration cycles are unlimited-feeling without burning credits

💡 Use Cases

⚡Studios needing motion-style control across a campaign — set physics_v2 + better_motion once and re-use across many stills

⚡Spicy creator workflows where each clip needs a different motion signature — dial the LoRA stack per shot

⚡VFX previz where the motion has to match a specific reference clip's physics — dial physics_v2 to mimic the reference

⚡Editorial fashion B-roll where stylized rendering matters — dial omninft / synth / dreamly for the visual signature

⚡Long-form 15-20s narrative motion where the base preset drifts off the intended look — LoRA override fixes the drift

⚡A/B testing motion styles at the same prompt — same image, same prompt, different LoRA objects, side-by-side outputs

⚡Power users coming from the base LTX 2.3 Spicy who need finer style control without leaving the LTX backbone

🎯 Best For

🎯 Power users, motion-design studios, spicy creators with a specific style reference, VFX teams and campaign producers who want unrestricted-feeling per-LoRA control over an i2v generation while keeping the platform's safety constraints intact.

👍 Pros

✓Per-LoRA strength override — the only spicy i2v on the catalog that exposes this level of motion-style control

✓Same 20s × 1080p ceiling as the base LTX 2.3 Spicy — duration is not compromised by the LoRA controls

✓5-second billing minimum + 480p preview mode = unlimited-feeling LoRA experimentation at low cost

✓Browse the model page with no login required; once logged in, session stays cached for no login mid-workflow

✓Backward-compatible with the base model — leave LoRAs blank and you get base behavior at LoRA pricing

⚠️ Considerations

△Slightly higher base price than the non-LoRA variant — switch back to <a href="/model/ltx-2-3-spicy-image-to-video">LTX 2.3 Spicy Image to Video</a> if you don't need LoRA control

△LoRA JSON requires correct syntax — malformed JSON falls back to preset defaults silently

△Learning curve on which LoRA names map to which visual behaviors — see Pro Tips below

△Audio is not generated — output is video-only; pair with a separate TTS or score model for final cuts

📚 How to Use LTX 2.3 Spicy Image to Video LoRA

Upload your reference image — JPG or PNG only, clear subject and good lighting

Write a one or two sentence motion / camera / mood prompt

Pick a preset — 'tuned' is the recommended default; 'original' is a lighter stack

(Optional but the whole point) Fill the LoRA strength override field with JSON, e.g. {"physics_v2": 0.8, "better_motion": 0.6}

Pick resolution (480p iteration / 720p balanced / 1080p best) and duration (3-20s)

Generate, evaluate, tweak the LoRA strengths, re-run — unlimited iteration at 480p is cheap

💡 Pro Tips for LTX 2.3 Spicy Image to Video LoRA

★

Start with one LoRA at strength 0.6, then add a second Adding three or four LoRAs at strength 1.0 at the same time produces chaotic outputs — they fight each other. The disciplined workflow: pick the one LoRA closest to your target look (physics_v2 for grounded motion, better_motion for smoother limbs, omninft for stylized rendering), run at strength 0.6, evaluate, then layer a second LoRA at 0.3-0.4. Iterating two LoRAs at moderate strengths consistently beats stacking five at full strength.

★

Use the same image with different LoRA objects to A/B test style Because the base motion behavior is deterministic at the same seed, you can run the same image + prompt + seed with two different LoRA objects and the only thing that changes is the motion style. This is the fastest way to learn what each named LoRA actually does. Run a 5-second 480p test grid (cheap), pick the winner, scale up to 1080p × 15-20s for the final.

★

physics_v2 + better_motion is the safest default for long clips If you're picking just two LoRAs to dial up on a 15-20 second generation, physics_v2 (real-world mass and momentum) plus better_motion (smoother limb / fabric interpolation) is the combo that holds up best across long durations. Strengths around 0.7 and 0.5 respectively. This is roughly equivalent to the base LTX 2.3 Spicy behavior but with stronger motion realism at the long-clip end.

★

Disable specific LoRAs inside a preset by setting them to 0 The preset (tuned / original) ships with a default set of LoRAs already active. If the preset is mostly right but one specific LoRA is producing an unwanted effect — for example you want tuned but without hardcut transitions — explicitly set {"hardcut": 0} in the LoRA object. The preset's other LoRAs stay, only hardcut is removed. This is subtractive control that no other model in the spicy family offers.

Ready to try LTX 2.3 Spicy Image to Video LoRA?

Get 10 free credits — no credit card required

Start Free →

Frequently Asked Questions

Browsing requires no login and no sign up. You can read the model description, watch examples and inspect pricing without an account. Only the generation step requires being logged in so we can attribute credits and the output to you. If your account says no credits, you have used the complimentary starter pack — top up from Billing and you can keep running unlimited LoRA experiments without restrictions.

It's a flat object mapping LoRA name to a number between 0 and 1. Example: {"physics_v2": 0.8, "better_motion": 0.6, "omninft": 0.4}. Available LoRA keys: sulphur, sulphur_v1, vbvr, dreamly, synth, plora, singularity, omninft, omninft_bf16, better_motion, physics_v2, hardcut, transition. Unknown keys are silently ignored. Set a key to 0 to disable that LoRA inside your chosen preset. Malformed JSON falls back to the preset's default stack with no error — so if you tweak the JSON and nothing changes, the JSON is probably invalid.

Use the base LTX 2.3 Spicy Image to Video when the tuned preset already gives you the look you want — it's cheaper and faster. Use this LoRA variant when the base keeps producing the same flavor of motion and you need to push it elsewhere, OR when you are running a campaign where motion style needs to be locked across many generations.

20 seconds at 1080p — the same as the base LTX 2.3 Spicy. Duration is billed at max(5, min(duration, 20)) seconds, so 3s and 4s clips cost the same as 5s, and the model rejects anything over 20s upstream. 720p costs about 1.67× the 480p price; 1080p costs about 2.33×.

No. People search for terms like 'ai image to video uncensored', 'ai image to video generator uncensored' and 'unrestricted ai video generator' but JAI Portal is strictly a B2B tool that does NOT support uncensored, nsfw or otherwise unrestricted explicit image/video generation. If an upstream model technically allows uncensored output, our platform is not responsible for that output. We run strict moderation, face-swap safeguards and content filters to prevent generations that are nsfw, illegal or considered similar. For unrestricted creative motion within those guidelines (cinematic, fashion, dance, action, glamour shots) the model performs well — for explicit content we cannot help and we do not lift the restrictions.

Approximate behavior (informal — actual effect varies by image and prompt): physics_v2 = grounded mass + momentum; better_motion = smoother limb / fabric interpolation; omninft / omninft_bf16 = stylized rendering bias; synth / dreamly = dreamlike soft style; plora = portrait-friendly skin and lighting; singularity = strong subject lock across the clip; vbvr / sulphur / sulphur_v1 = specialised motion styles; hardcut = forces cut-style transitions; transition = smooth transition behavior. The honest workflow is to test each one in isolation at strength 0.6 on the same image + prompt + seed to learn its signature, then combine. Unknown keys are ignored so experimenting carries no risk.

If you are going to run more than 3-4 generations on the same image / campaign, yes — the LoRA control lets you converge on the right look faster and that saves more credits than the per-generation surcharge costs. If you are running a single one-off generation and the base 'tuned' preset already gives you the look you want, the base LTX 2.3 Spicy Image to Video is cheaper and equally fast. As a rule of thumb: production campaigns → LoRA variant; quick one-off creative runs → base model.

Yes. JAI Portal exposes every catalog model including this LoRA variant via the standard generation API. The LoRA JSON object passes straight through to the upstream wavespeed endpoint. Your team can run unlimited batch generations programmatically — set the LoRA stack once per campaign config, swap reference images and prompts, and ship. There is no API rate limit beyond the standard per-IP and per-user limits, and there is no separate 'unrestricted' API tier — the same content moderation that applies to UI generations applies to API generations. Contact support for enterprise volume pricing if you are running thousands of generations per day.

If a generation is blocked, the model or our pre-flight moderation flagged either the input image or the prompt for content that violates platform policy. There is no unrestricted mode — JAI Portal does not ship an unfiltered, 'no restrictions', 'ai image to video uncensored' or 'ai image to video generator uncensored' variant for any spicy model, including this one. The long-tail keywords 'uncensored ai video', 'unrestricted ai video generator', 'no restrictions image to video' and 'nsfw image to video' are commonly searched but we deliberately do not advertise or enable that behavior. If your input is artistic / fashion / dance / cinematic and you still see a block, soften the prompt language, try a different reference image, or contact support so we can review the false-positive. We accept that some traffic looking for nsfw/uncensored tools will bounce — protecting the platform is more important than catching that traffic.

⚖️ How LTX 2.3 Spicy Image to Video LoRA Compares

LTX 2.3 Spicy Image to Video LoRA is the most flexible spicy image-to-video model on JAI Portal. It takes the same 20-second 1080p backbone as the base LTX 2.3 Spicy Image to Video and adds full per-LoRA strength control via a JSON object — slap, choke, physics, motion quality, transitions, stylization, all individually dial-able from 0 to 1. Important framing for users arriving from searches like 'ai image to video LoRA uncensored' or 'ai image to video generator LoRA uncensored': the LoRAs here control MOTION style only — they do NOT unlock uncensored or unrestricted explicit output. JAI Portal does not ship an uncensored variant of any spicy model and the LoRA stack is intentionally limited to motion / style behaviors. For finding the right baseline style: pick this LoRA variant. For fast one-off creative runs that just need the default 'tuned' look: pick the base model. For workflow-style spicy creation with pre-baked single-motion presets like slap or choke, the dedicated JAI Slap LoRA, JAI Choke LoRA and JAI Pose LoRA workflows on the WAN backbone are simpler one-tap options — pair them with this model when the campaign has both a known action shot and an open cinematic shot. For sibling spicy options on different backbones, WAN 2.2 Spicy Image to Video is the best 5-8s fast iteration option, Vidu Q3 Image to Video Spicy is the cheapest short-clip generator, and Bytedance Seedance v1.5 Pro Image to Video Spicy is the option to reach for when you need native audio. There is no unrestricted / uncensored / nsfw mode on the LoRA variant or on any spicy model on JAI Portal — every model on the platform runs the same moderation pipeline and the keywords 'unrestricted' and 'no restrictions' refer to the iteration workflow (unlimited credit-bounded runs, no sign up to browse, no login prompts mid-workflow once logged in), not to content filters. The best AI image-to-video workflow on the platform is: iterate prompts at 480p with this LoRA variant, lock the LoRA object, then run the final at 1080p × 15-20s for the keeper.

LTX 2.3 Spicy Image to Video LoRA

More Video Generation Models