LTX 2.3 vs 2.5 on a DGX Spark: the older model rendered in two-thirds the time and opened the drawer 2.5 didn't

Rated G

Same prompts, seeds and settings on one Spark. LTX 2.5 still made the cleaner dog and the camera glide, so if you have 2.3, keep both.

Instructional (150 s): AI hosts Daisy Lane and Rafael Duarte (AI voices, ElevenLabs) put LTX 2.3 and LTX 2.5 side by side on the same three prompts, show what each costs on a Spark, and say which to use for what.
Prompt and settings

Settings: Four host segments (52 s of speech, ElevenLabs eleven_v4) made with LTX 2.5 image+audio-to-video from each host's on-set still, seed 42; five voice-over panels in Remotion built from this post's own clips; music bed ElevenLabs Music (AI-generated); 1920x1080, 30 fps

Reference images (2)

The dog: LTX 2.3 (left) vs LTX 2.5

  • Clip 2: LTX 2.3: the dog looks photographed, with real kitchen light. But the coat is rough when we asked for smooth, and after holding the look for about three seconds it turns back to profile at 4.3 s, unasked.
    Prompt and settings

    Prompt: Side view of a small smooth-coated brown and tan miniature dachshund standing still on a light wood kitchen floor in a bright sunlit kitchen, long body in profile filling the middle of the frame, it turns its head to look straight at the camera with a proud serious expression and blinks. Almost motionless. The camera is almost still. Full-frame image filling the entire frame edge to edge, photorealistic, warm morning light.

    Settings: LTX 2.3 22B dev fp8 + distilled LoRA 384 at 0.5, text-to-video, 1920x1088, 121 frames at 24 fps (5.04 s), seed 2207, 8 + 3 steps, euler, CFG 1.0. Rendered Nov 19 2026, 149.9 s cold. Audio track removed for this page.

  • Clip 3: LTX 2.5: the dachshund we described turns to camera and holds to the end. It looks like a polished 3D render more than a photo.
    Prompt and settings

    Prompt: Side view of a small smooth-coated brown and tan miniature dachshund standing still on a light wood kitchen floor in a bright sunlit kitchen, long body in profile filling the middle of the frame, it turns its head to look straight at the camera with a proud serious expression and blinks. Almost motionless. The camera is almost still. Full-frame image filling the entire frame edge to edge, photorealistic, warm morning light.

    Settings: LTX 2.5 22B distilled int8, text-to-video, 1920x1088, 121 frames at 24 fps (5.04 s), seed 2207, 8 + 3 steps, euler ancestral, CFG 1.0. Rendered Oct 31 2026; re-rendered Nov 19 identical, 183.2 s on Oct 31.

The pantry glide: LTX 2.3 vs LTX 2.5

  • Clip 4: LTX 2.3: drifts down and a little sideways in the first second and a half, then stops dead. Not the slow glide we asked for. Labels are crisp letters that spell nothing (OALINS, DENNTS).
    Prompt and settings

    Prompt: A beautifully organized pantry with light oak shelves holding rows of matching clear bins and glass jars of pasta, rice, oats, lentils and snacks, each with a small neat white rectangular label. Warm morning sunlight streams in. The camera glides very slowly sideways along the shelves. Full-frame image filling the entire frame edge to edge, photorealistic home lifestyle commercial, cream and sage palette.

    Settings: LTX 2.3 22B dev fp8 + distilled LoRA 384 at 0.5, text-to-video, 1920x1088, 97 frames at 24 fps (4.04 s), seed 909, 8 + 3 steps, euler, CFG 1.0. Rendered Nov 19 2026, 91.3 s warm. Audio track removed for this page.

  • Clip 5: LTX 2.5: the slow sideways glide we asked for, with parallax between the bins and the posts. Labels are smudged pseudo-text.
    Prompt and settings

    Prompt: A beautifully organized pantry with light oak shelves holding rows of matching clear bins and glass jars of pasta, rice, oats, lentils and snacks, each with a small neat white rectangular label. Warm morning sunlight streams in. The camera glides very slowly sideways along the shelves. Full-frame image filling the entire frame edge to edge, photorealistic home lifestyle commercial, cream and sage palette.

    Settings: LTX 2.5 22B distilled int8, text-to-video, 1920x1088, 97 frames at 24 fps (4.04 s), seed 909, 8 + 3 steps, euler ancestral, CFG 1.0. Rendered Oct 31 2026; re-rendered Nov 19 identical, 163.0 s cold on Nov 19.

The freezer drawer: LTX 2.3 vs LTX 2.5

  • Clip 6: LTX 2.3: the drawer comes toward the camera in the first second and lands on a full drawer of bags. A bit of a lurch as the drawer front drops out of frame, but the move happened.
    Prompt and settings

    Prompt: A freezer drawer slides open toward the camera revealing perfectly organized rows of flat clear freezer bags of portioned chicken, ground beef and soup, each with a small neat white rectangular sticker label. Soft cold light with a little frosty mist drifting out, clean and satisfying. Full-frame image filling the entire frame edge to edge, photorealistic product commercial.

    Settings: LTX 2.3 22B dev fp8 + distilled LoRA 384 at 0.5, text-to-video, 1920x1088, 97 frames at 24 fps (4.04 s), seed 808, 8 + 3 steps, euler, CFG 1.0. Rendered Nov 19 2026, 91.2 s warm. Audio track removed for this page.

  • Clip 7: LTX 2.5: a slow drift over packages. No drawer, no slide.
    Prompt and settings

    Prompt: A freezer drawer slides open toward the camera revealing perfectly organized rows of flat clear freezer bags of portioned chicken, ground beef and soup, each with a small neat white rectangular sticker label. Soft cold light with a little frosty mist drifting out, clean and satisfying. Full-frame image filling the entire frame edge to edge, photorealistic product commercial.

    Settings: LTX 2.5 22B distilled int8, text-to-video, 1920x1088, 97 frames at 24 fps (4.04 s), seed 808, 8 + 3 steps, euler ancestral, CFG 1.0. Rendered Oct 31 2026; re-rendered Nov 19 identical, 139.8 s warm on Nov 19.

Verdict

LTX 2.5 isn't a clean upgrade over LTX 2.3 on a Spark: 2.3 finished the same 4-second 1920x1088 clip in 91 s against 140 s, and it pulled open the freezer drawer that 2.5 ignored. 2.5 still won the other two: it did the sideways camera glide 2.3 skipped, and its dachshund held the pose while 2.3's turned back to profile. 2.3 looks more like real footage; 2.5 looks more like a clean render. That's three prompts with one seed each, so treat it as a lean: if 2.3 is on your box, keep it, and try both on any shot where something has to move.

Credits and settings

ModelLTX 2.3 22B dev vs LTX 2.5 22B distilled (text-to-video)
MakerLightricks
VersionLTX 2.3: ltx-2.3-22b-dev-fp8 + ltx-2.3-22b-distilled-lora-384 (0.5), Gemma 3 12B fp4. LTX 2.5: 22B distilled int8 convrot, Gemma 4 12B int8
What it doestext-to-video
LicenseLTX-2 Community License Agreement (Jan 5, 2026); free under $10M annual revenue
HardwareDGX Spark: NVIDIA GB10, ARM (aarch64), 128 GB unified memory. ComfyUI 0.43.0, PyTorch 2.11.0+cu130, no custom nodes. LTX 2.3 fp8 checkpoint; LTX 2.5 int8 transformer.
Speed23 s of compute per second of video. LTX 2.3, 1920x1088 text-to-video, 24 fps, two stages (8 steps at 960x544, x2 latent upscale, 3 steps), weights loaded: 91.3 s for 4.04 s. LTX 2.5 on the same graph and machine: 34.6.
Peak memory43.5 GB. ComfyUI process peak from nvidia-smi once a second: 41,509 MiB (40.5 GiB) for LTX 2.3, 41,597 MiB for LTX 2.5. Per-process share of the GB10 unified memory, not a machine total.
Tested onNov 19, 2026
ReleasedMar 5, 2026
AI hosts

Daisy Lane (AI host; AI voice (ElevenLabs), AI voice)

Rafael Duarte (AI host; AI voice (ElevenLabs), AI voice)

Narration
Not supplied
Music source
ElevenLabs Music, AI-generated

What it does well

  • About 35% less time per clip: 91 s warm for 4.04 s at 1920x1088 (22.6 s per second of video) against 140 s for LTX 2.5 on the same machine the same morning.
  • Did the big move: the freezer drawer slides toward the camera in the first second, where LTX 2.5 gave a static drift.
  • Looks photographed: real light falloff, textures and kitchen clutter in the dog shot.
  • Same memory as LTX 2.5: 40.5 GiB process peak against 40.6 GiB.
  • Runs in stock ComfyUI 0.43.0 on aarch64 with the official template's nodes, no patches.

Where it breaks

  • Missed the camera glide in the pantry prompt: a quick drift down and sideways, then it stops dead (clip 3).
  • Adds motion you didn't ask for: the dog turns back to profile at 4.3 s (clip 1).
  • Drifts from the description: a rough-coated dog when the prompt says smooth-coated (clip 1).
  • Can't write: labels are sharp letters that spell nothing (clip 3).
  • The drawer move lurches: the drawer front drops out the bottom of frame as the camera seems to dive in (clip 5).
  • Bigger download than LTX 2.5: about 47 GB for the 2.3 set against about 39 GB.

How we tested

Three prompts from our LTX 2.5 product-shot batch, rendered again with LTX 2.3 on the same Spark. Same text, same seeds, same 1920x1088 size, same frame counts, same two-stage sigma schedule. The only things we swapped were the model files and the sampler that each model's own ComfyUI template uses. The LTX 2.5 clips on this page are the ones from our [LTX 2.5 review](/video-lab/ltx-2-5-22b-distilled-text-to-video).

We picked those three because each one asks for something different:

  • The dachshund asks for one small move: turn the head to camera and blink. LTX 2.5 nailed it, so it's the bar.
  • The pantry asks for a slow sideways camera glide. LTX 2.5 did that one too.
  • The freezer asks for the big move: a drawer sliding open toward the camera. LTX 2.5 ignored it.

To make the speed comparison fair, we also re-ran LTX 2.5 on two of the prompts the same morning, on the same idle machine. Those re-runs came out frame-for-frame identical to the October clips, so the clips on the page and today's timings belong together.

The clips are above, each pair in a side-by-side player. One thing the stills can't show: LTX 2.3's drawer move lurches a little. The drawer front drops out of the bottom of the frame as the camera seems to dive in. The move still happened, and on a product shot that's the part you can't fake.

Setup and gotchas

  • Nothing new to download if 2.3 is already on your box. We ran it in the same ComfyUI 0.43.0 as LTX 2.5, with no custom nodes. The graph follows ComfyUI's own LTX-2.3 text-to-video template: the 2.3 checkpoint (it carries the video VAE and audio VAE inside it), the distilled LoRA on top at strength 0.5, the Gemma 3 12B text encoder, and the 2.3 x2 latent upscaler.
  • Our distilled LoRA is the older one. The template names version 1.1 of the distilled LoRA. We had the original ltx-2.3-22b-distilled-lora-384 on disk and used it at the template's 0.5. It worked fine, but if you're starting fresh, get 1.1.
  • Starting from scratch, 2.3 is the bigger download. The 2.3 set is about 47 GB: the fp8 dev checkpoint (29.1 GB), the distilled LoRA (7.6 GB), Gemma 3 12B fp4 (9.4 GB) and the upscaler (1.0 GB). LTX 2.5's set is about 39 GB.
  • Different text encoders. LTX 2.3 reads prompts with Gemma 3 12B, LTX 2.5 with Gemma 4 12B. You can't share one file between them.
  • 2.3's template decodes sound; our 2.5 graph didn't. The 2.3 clips came out with an audio track. We stripped it on this page so you're comparing pictures with pictures.
  • The negative prompt does nothing on either. Both models run at CFG 1, so the negative prompt has no effect either way.

Settings

Text-to-video, 1920x1088, 24 fps. The dog is 121 frames (5.04 s); the pantry and freezer are 97 frames (4.04 s). Two stages: stage one samples at 960x544 with 8 fixed sigmas (1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0), then the x2 latent upscaler, then 3 more sigmas (0.85, 0.725, 0.4219, 0.0) at full size. Stage-one seeds 2207 (dog), 909 (pantry), 808 (freezer); stage-two seed 42. Tiled VAE decode (tile 512, overlap 64, temporal 64 / 16).

LTX 2.3LTX 2.5
Model fileltx-2.3-22b-dev-fp8 + distilled LoRA 384 at 0.5ltx-2.5-22b-distilled-transformer-comfy-int8-convrot
Text encoderGemma 3 12B fp4 mixedGemma 4 12B int8
Sampler and guidereuler, CFG guider 1.0euler ancestral, LTXV dual-CFG guider 1.0 / 1.0
Upscalerltx-2.3-spatial-upscaler-x2-1.1ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0

The workflow file linked below is the exact API-format graph we sent for the 2.3 pantry clip.

Speed and memory

Wall time is from sending the graph to ComfyUI until it reported the clip done. It covers both sampling stages, the latent upscale, the VAE decode and the mp4 write. "Cold" means we had just told ComfyUI to unload everything, so the clip paid for loading the weights.

ClipModelWall timeCompute per second of video
Pantry, 4.04 s, warmLTX 2.391.3 s22.6 s
Freezer, 4.04 s, warmLTX 2.391.2 s22.6 s
Dog, 5.04 s, coldLTX 2.3149.9 s29.7 s
Freezer, 4.04 s, warmLTX 2.5139.8 s34.6 s
Pantry, 4.04 s, coldLTX 2.5163.0 s40.3 s

Warm, LTX 2.3 finished the same 4-second clip in 91 s against 140 s for LTX 2.5. That's about 35% less time, on the same machine, the same morning, with nothing else on the GPU. We didn't break down where the difference comes from (fp8 checkpoint versus int8 transformer, the text encoder, the sampler), so take it as the speed of each setup as shipped.

Memory came out the same. The ComfyUI process peaked at 40.5 GiB with LTX 2.3 and 40.6 GiB with LTX 2.5, read from nvidia-smi once a second. On the GB10's unified memory that's this process's share, not a machine total. The lowest free memory we saw on the box during a run was 51 GiB with 2.3 and 46 GiB with 2.5.

What to do with this

If LTX 2.3 is already on your Spark, keep it. Reach for it when you want the shot to look like footage, when a prop has to move, or when you're burning through seeds and time matters. Reach for LTX 2.5 when the subject has to look clean and stay put, or when the move is a camera glide. Either way, don't trust either one with readable text: put labels in afterwards.

ComfyUI workflow for the LTX 2.3 pantry clip (API format) ↗