--- title: "LTX 2.3 Fast: Specs, 4K Output & Prompt Tips | imaginary.app" description: "LTX 2.3 Fast is the speed tier of Lightricks’ open LTX-2.3 video model: up to 2160p, optional sound, portrait support. How it compares to Pro, plus tips." url: https://www.imaginary.app/models/ltx-2-3 --- Updated November 2026 # LTX 2.3 Fast LTX 2.3 Fast is the speed-optimised variant of LTX-2.3, Lightricks’ open-weights audio-and-video model, and it can output 1080p, 1440p or 2160p with optional sound generated in the same pass. Lightricks positions Fast for quick iteration and the Pro variant for maximum fidelity; on imaginary.app we run the Fast variant. Key takeaways - This page covers LTX-2.3 Fast (fal endpoint fal-ai/ltx-2.3/text-to-video/fast), not LTX-2.3 Pro or the newer LTX-2.5. - Lightricks announced LTX-2.3 on February 23, 2026, with a rebuilt VAE for sharper detail, better prompt understanding, cleaner audio and native portrait video. - Lightricks says Fast prioritises speed while Pro delivers sharper detail, truer colour and smoother motion. - Lightricks released LTX-2.5, the successor to LTX-2.3, on August 11, 2026. We currently offer 2.3 Fast. Lightricks Generate with LTX 2.3 Fast Write a prompt, pick LTX 2.3 Fast, choose your settings and see the credit cost before you start. Failed generations return their credits. [Open the generator](https://www.imaginary.app/generate)[View Pricing](https://www.imaginary.app/pricing) Sign in with Google to create an account; buy credits only when you need them. ## On imaginary.app These are the settings our generator offers for LTX 2.3 Fast right now, read from our live model catalog. - Type Video, from a text prompt - Resolutions 1080p, 1440p, 2160p - Aspect ratios 16:9, 9:16 - Clip lengths 6 s, 10 s - Audio Optional: switch generated sound on or off - Starting price 2.93 credits (about $0.62–$0.73) at the lowest settings (1080p, 16:9, 6 s, sound off) The generator shows the exact credits for your settings before you start. USD ranges depend on which [pack or plan](https://www.imaginary.app/pricing) you bought credits with. If a generation fails, or you cancel it before it starts, its credits go back to your balance. Very rarely, when we can't confirm with the provider whether a generation finished, the credits stay reserved while we check, rather than being charged. Models other than Veo 3.1 need an account that has bought a credit pack or plan, and they have daily usage limits, per account and across the site: if one is reached, the generator tells you and you can try again the next day (limits reset at midnight UTC). ## Fast vs Pro: same family, different priorities Lightricks’ API ships LTX-2.3 in two variants. Its March 5, 2026 changelog says Pro supports every endpoint, including audio-to-video, retake and extend, while Fast supports text-to-video and image-to-video. Lightricks’ own guide puts the trade-off plainly: Fast mode prioritises speed, and Pro delivers higher fidelity, with sharper detail, truer colour and smoother motion, recommended for final delivery and hero shots. fal describes its Fast endpoint as the same model with faster output for rapid iteration. ## High output resolution, including vertical fal’s Fast endpoint offers 1080p, 1440p and 2160p, in 16:9 or 9:16. Lightricks says 2.3 is the first LTX release with native portrait video, up to 1080×1920, trained on vertical footage rather than cropped from landscape. ## Sharper detail from a rebuilt VAE Lightricks rebuilt the model’s VAE and latent space for LTX-2.3. It says fine textures such as hair and edges were softer than intended in earlier checkpoints, and that 2.3 is sharper at all resolutions. ## Audio and video in one model LTX-2.3 generates sound and picture together: Lightricks describes a dual-stream design in which the audio and video streams share transformer blocks and condition each other during generation. For 2.3 it filtered the training data for silence, noise and artifacts and shipped a new vocoder, which it says gives fewer random sounds and tighter alignment. The model card adds one caution: audio without speech may be lower quality. ## LTX 2.3 Fast specs from Lightricks What Lightricks and its API partners document for the model itself. A maker spec is not always a setting our generator offers: what you can choose on imaginary.app is listed in the block above. | Spec | As documented | Source | | ---------------------------------- | ---------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | | Maker | Lightricks | [Lightricks — “LTX-2.3 and LTX Desktop” release post (February 23, 2026)](https://ltx.io/model/model-blog/ltx-2-3-release) | | Release | LTX-2.3 announced February 23, 2026 | [Lightricks — “LTX-2.3 and LTX Desktop” release post (February 23, 2026)](https://ltx.io/model/model-blog/ltx-2-3-release) | | API variants | Pro and Fast listed in the LTX API changelog on March 5, 2026 | [LTX API changelog — March 5, 2026 (LTX-2.3 Pro and Fast)](https://docs.ltx.io/api-changelog/2026/3/5) | | Variant on this page | Fast: speed-optimised text-to-video (fal-ai/ltx-2.3/text-to-video/fast) | [fal — LTX 2.3 Fast text-to-video](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast) | | Clip length on fal’s Fast endpoint | 6 to 20 seconds; clips over 10 seconds require 1080p at 25 fps | [fal — LTX 2.3 Fast text-to-video schema (llms.txt)](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast/llms.txt) | | Resolutions on fal’s Fast endpoint | 1080p, 1440p, 2160p | [fal — LTX 2.3 Fast text-to-video schema (llms.txt)](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast/llms.txt) | | Aspect ratios | 16:9 and 9:16; native portrait up to 1080×1920 | [Lightricks — “LTX-2.3 and LTX Desktop” release post (February 23, 2026)](https://ltx.io/model/model-blog/ltx-2-3-release) | | Frame rates on fal’s Fast endpoint | 24, 25, 48 or 50 fps | [fal — LTX 2.3 Fast text-to-video schema (llms.txt)](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast/llms.txt) | | Audio | Optional sound generated with the video | [fal — LTX 2.3 Fast text-to-video schema (llms.txt)](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast/llms.txt) | | Weights | Open weights on Hugging Face under the LTX-2 Community License | [Hugging Face — Lightricks/LTX-2.3 model card](https://huggingface.co/Lightricks/LTX-2.3) | | Licence terms | Companies above a stated annual-revenue threshold need a separate commercial licence from Lightricks | [Lightricks — LTX-2 Community License (LICENSE-2)](https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2) | | Successor | LTX-2.5, released August 11, 2026 | [Lightricks — LTX release notes (LTX-2.5, August 11, 2026)](https://ltx.io/release-notes) | ## Who LTX 2.3 Fast is for ### Good fit - High-resolution B-roll and establishing shots where a 1440p or 2160p file is useful in the edit. - Native vertical clips for Reels, Shorts and TikTok that need to look sharp on a phone. - Short scenes with a spoken line, when you direct the delivery beat by beat. ### Look elsewhere if - Hero shots where every detail counts. Lightricks itself recommends Pro for final delivery, and LTX-2.5 is now its newest model. - Square or ultra-wide formats: the Fast endpoint offers only 16:9 and 9:16. - Pure soundscapes or music without speech, which the model card flags as lower quality. - Audio-to-video, retake or extend workflows, which are Pro features on Lightricks’ API and not offered on imaginary.app. ## Prompt tips for LTX 2.3 Fast Write long, specific prompts. Lightricks says longer, more descriptive prompts consistently outperform short ones on 2.3, and that if you had been simplifying prompts for older checkpoints you should be more specific now. Cover camera movement, lighting, subject action and visual style. Example prompt (written by us) ``` Slow drone descent into a misty pine valley at dawn, a single wooden cabin with smoke rising from its chimney, soft golden light breaking through the fog, 24mm lens, gentle forward movement, muted greens and warm highlights, documentary style. Sound: birdsong, a distant stream, wind moving through the trees. ``` For dialogue, split the line into short phrases with one acting cue between each. Lightricks’ dialogue guide says 2.3 reads direction at the phrase level and that long unbroken sentences make the timing collapse. Example prompt (written by us) ``` A young woman in a rain-streaked car at night speaks quietly, "I almost didn’t come..." She pauses and looks out of the window, then continues, "but I had to see it." A faint smile fades. The camera holds a close-up from the passenger seat. The audio is intimate, with rain on the roof and a low engine idle. ``` Describe sounds by material and space. Lightricks suggests audio language such as “echoing footsteps on stone” or “crackling fire with distant wind” rather than just naming a sound. Example prompt (written by us) ``` Vertical close-up of a blacksmith’s hammer striking glowing steel on an anvil, sparks spraying towards the lens, dim forge lit by orange firelight. Sound: sharp metallic ringing in a small stone workshop, the hiss of steel dipped in water, a roaring furnace behind. ``` Use slow, continuous camera moves for longer clips. Lightricks suggests a slow dolly so a longer shot reads as intentional. ## Limitations - LTX 2.3 Fast is made by Lightricks; imaginary.app is not affiliated with Lightricks. - This is the Fast variant. Lightricks says Pro gives sharper detail, truer colour and smoother motion, and LTX-2.5 has since succeeded 2.3. - On imaginary.app LTX 2.3 Fast runs text-to-video only, so image-to-video and first/last-frame interpolation are not offered here. - The model card notes that prompt following depends heavily on prompting style and that outputs may not match prompts perfectly. - Audio without speech may be lower quality, according to the model card. - Licences differ by part: Lightricks’ release notes describe the code and LTX Desktop as Apache 2.0, while the weights are under the LTX-2 Community License (fal’s model page calls the model Apache 2.0). We follow the weights’ licence. ## Related models - [Veo 3.1 Fast](https://www.imaginary.app/models/veo-3-1-fast) - [PixVerse V6](https://www.imaginary.app/models/pixverse-v6) - [Wan 3.0](https://www.imaginary.app/models/wan-3-0) - [Seedance 2.5](https://www.imaginary.app/models/seedance-2-5) - [Every model on imaginary.app](https://www.imaginary.app/models) - [AI video generator overview](https://www.imaginary.app/ai-video-generator) ## Frequently asked questions ### LTX 2.3 Fast vs Pro: what is the difference? Both are variants of LTX-2.3. Lightricks says Fast prioritises speed and Pro delivers higher fidelity, with sharper detail, truer colour and smoother motion. On Lightricks’ API, Pro also supports audio-to-video, retake and extend. imaginary.app offers the Fast variant. ### Is LTX 2.3 the latest LTX model? No. Lightricks released LTX-2.5 on August 11, 2026, adding native multishot, a new diffusion video decoder and a custom Gemma 4 12B text encoder. This page covers LTX-2.3 Fast, which is the version we offer. ### Can LTX 2.3 Fast make 4K video? fal’s Fast endpoint offers up to 2160p. The resolutions and durations we offer are listed in the “On imaginary.app” block on this page. ### Does LTX 2.3 have audio? Yes, optionally. LTX-2.3 generates sound and picture together in one model. Dialogue and sound effects tend to work better than speech-free audio, which the model card flags as potentially lower quality. ### Is LTX 2.3 open source? Its weights are published on Hugging Face under the LTX-2 Community License, and Lightricks requires a separate commercial licence for companies above an annual-revenue threshold it sets. On imaginary.app you use the hosted Fast variant through fal, so you don’t need to run the weights yourself. ### Can I use LTX 2.3 Fast videos commercially? The LTX-2 Community License governs the weights, including the annual-revenue threshold above which Lightricks requires a separate commercial licence. On imaginary.app you generate through fal’s hosted endpoint rather than running the weights, so check Lightricks’ licence, fal’s terms and our Terms of Service for how they apply to your clips. ## Sources Prices, model specs and dates on this page were checked on November 10, 2026. Third-party prices change; check the source before you rely on a number. - [Lightricks — “LTX-2.3 and LTX Desktop” release post (February 23, 2026)](https://ltx.io/model/model-blog/ltx-2-3-release) - [LTX API changelog — March 5, 2026 (LTX-2.3 Pro and Fast)](https://docs.ltx.io/api-changelog/2026/3/5) - [Lightricks — How to generate 20-second AI videos with LTX-2.3](https://ltx.io/blog/how-to-generate-20-second-ai-videos) - [Lightricks — Directing dialogue and acting in LTX-2.3](https://ltx.io/blog/directing-dialogue-and-acting) - [Lightricks — A guide to LTX-2.3 audio](https://ltx.io/blog/a-guide-to-ltx-2-3-audio) - [Lightricks — LTX release notes (LTX-2.5, August 11, 2026)](https://ltx.io/release-notes) - [Hugging Face — Lightricks/LTX-2.3 model card](https://huggingface.co/Lightricks/LTX-2.3) - [Lightricks — LTX-2 Community License (LICENSE-2)](https://github.com/Lightricks/LTX-2/blob/main/LICENSE-2) - [fal — LTX 2.3 Fast text-to-video](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast) - [fal — LTX 2.3 Fast text-to-video schema (llms.txt)](https://fal.ai/models/fal-ai/ltx-2.3/text-to-video/fast/llms.txt) - [imaginary.app pricing](https://www.imaginary.app/pricing) How we checked: model specs come from Lightricks's and its API partners' pages listed above; the imaginary.app settings and starting price come from our live model catalog. Example prompts are ours, not Lightricks's. Commercial use: licence and usage terms for LTX 2.3 Fast are set by Lightricks. Check them, and our [Terms of Service](https://www.imaginary.app/terms-of-service), before you publish what you make. Lightricks ## Try LTX 2.3 Fast on your own prompt One account and one credit balance for LTX 2.3 Fast, Veo 3.1 and every other video model we offer. The price shows before you generate. [Open the generator](https://www.imaginary.app/generate)[View Pricing](https://www.imaginary.app/pricing) Sign in with Google to create an account; buy credits only when you need them.