Read 2026-11-11 · Nano Banana 2.1 · 1K · 8
LTX 2.5 Upscale is a 2x LoRA note. This form is Nano Banana 2.1
The card says a 2x generative upscaler. docs.ltx.io also says 2× or 4×. This button sends a 1K still, not that LoRA.
This form is Nano Banana 2.1 at 1K for 8 credits. It does not run the LTX 2.5 Upscale LoRA, download weights, or return an upscaled video. 16 credits are 2 stills.
16 free credits = 2 standard images. No card required.
Output board


Your generated variations will appear here.
- Made by
- Lightricks
- Run on Nano AI
- Nano Banana 2.1 · 8 credits per image
- Page updated
- November 11, 2026
Source: huggingface.co
Photographs, not LTX 2.5 Upscale outputs
Alemily, CC BY-SA 2.5, original upload date 6 July 2006. The chart reads RESOLVING POWER TEST TARGET, USAF 1951, and BUCKBEE MEARS CO. Ixnayonthetimmay, public domain (PD-self), 2008-10-12. The Commons description names a Nikon Coolpix S610; no Nikon word is readable in the frame. Not LTX frames and not Nano Banana 2.1 outputs.


What LTX 2.5 Upscale publishes, and what this button sends
Read on 2026-11-11. The Hugging Face card for LTX 2.5 Upscale says it is a Pixel Spatial Upscaler IC-LoRA on LTX-2.5-22B, currently a 2x variant. The only weight file listed is ltx-2.5-22b-ic-lora-pixel-spatial-upscaler-x2-1.0.safetensors. The sentence on the card is a generative upscaler rather than a refiner. The repository tags include super-resolution. The card does not print the single phrase generative super-resolution. Rank 32, BF16, reference_downscale_factor 2, and LoRA strength 1.0 because the weights ship pre-scaled. The recommended draft is about 280p, then the 2x pass. The license name is ltx-2.x-community-license. The repo is gated behind Agree and Access. Hugging Face lists createdAt 2026-08-11 and lastModified 2026-10-22. The card itself does not print a separate launch date.
The usage guide on docs.ltx.io, read the same day, says this upscaler raises a clip by 2× or 4×, and that fidelity is tunable through LoRA strength, guidance, and step count. The adapters page on docs.ltx.io says separate 2× and 4× variants are provided, names those variants on LTX-2.3, and says the LTX-2.5 adapter is currently a 2× variant only. The card links LTX-2.5_V2V_ICLoRA_Single_Stage_Distilled.json. The adapters page says no dedicated LTX-2.5 workflow is linked there. The card's docs link uses docs.ltx.video. That host returned 403 on 2026-11-11, so the sentences here are from docs.ltx.io and are not a claim that the two hosts match.
A different file, ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors, is the spatial upscaler in the two-stage guide. That guide's heading is upscale-and-refine, and both stages use CFG 1. Native Resolution can apply that latent file twice on the 8K path. Refine Details is another LoRA, trained for a 1024×576 tile, and that page says not to substitute the Pixel Spatial Upscaler. The hosted API at https://api.ltx.io/v2/video-upscale lists 1080p at $0.08, 4K at $0.30, and 8K at $1.20 per second of input, and the short side must be at least 480 pixels. That page does not say the endpoint is this LoRA. This form is Nano Banana 2.1 at 1K for 8 credits. /ltx-2-5 stays the general guide and runs Veo 3.1 Fast for 60 credits. /unifab-video-upscaler-ai names Equinox, Vellum, Kairo, and Titanus. /relayvsr and /replicate-nightmareai are other upscaler notes, not this LoRA.
LTX 2.5 Upscale lines left side by side
- Card, read 2026-11-11
- 2x only · ltx-2.5-22b-ic-lora-pixel-spatial-upscaler-x2-1.0.safetensors · rank 32 · BF16 · reference_downscale_factor 2 · strength 1.0 · about 280p
- Words on that card
- generative upscaler rather than a refiner · tags include super-resolution · not the phrase generative super-resolution
- docs.ltx.io the same day
- 2× or 4× in the usage guide · LTX-2.5 currently 2× only on the adapters page · LTX-2.3 has 2× and 4× · docs.ltx.video returned 403
- Other Lightricks files
- ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors · upscale-and-refine · CFG 1 · 1024×576 · API 1080p $0.08, 4K $0.30, 8K $1.20 · short side 480
- Dates from the API
- createdAt 2026-08-11 · lastModified 2026-10-22 · license name ltx-2.x-community-license · gated Agree and Access
- This form
- Nano Banana 2.1 · 1K · 8 credits · 16 credits are 2 stills · not a video and not an upscale
Three upscale lines, and a still that is none of them
2x on the card, 2× or 4× in the usage guide
The card lists one file, the 2x checkpoint, and calls the result a generative upscaler. The tag list includes super-resolution. docs.ltx.io also says 2× or 4×, and then says the LTX-2.5 adapter is 2× only while LTX-2.3 has both.
Strength 1.0, and a 280p draft next to a 480 minimum
The card sets LoRA strength 1.0 and moves fidelity with guidance and step count. The usage guide also names LoRA strength as a fidelity control. The API short side is at least 480. Those lines are not merged here. Rank 32 and BF16 stay on the card.
This button, and the neighboring pages
Nano Banana 2.1, 1K, 8 credits. 16 credits are 2 stills. /ltx-2-5 is Veo 3.1 Fast at 60 credits. UniFab's Equinox, Vellum, Kairo, and Titanus stay on /unifab-video-upscaler-ai. /relayvsr and /replicate-nightmareai stay their own notes.
LTX 2.5 Upscale next to the other published lines
Read 2026-11-11. The card, the docs, and the API are not rewritten into one scale factor.
| Model card | docs.ltx.io and the API | This form | |
|---|---|---|---|
| File | ltx-2.5-22b-ic-lora-pixel-spatial-upscaler-x2-1.0.safetensors | ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors is a different file | no weight file |
| Scale | 2x only, generative upscaler, tag super-resolution | usage guide 2× or 4×; adapters page 2× only for LTX-2.5; API 1080p, 4K, 8K | 1K still |
| Control | strength 1.0, then guidance and step count; about 280p | usage guide also names LoRA strength; API short side at least 480; CFG 1 on the two-stage guide | no LoRA strength |
| Price | not sold here; ltx-2.x-community-license | $0.08, $0.30, $1.20 per second of input | 8 credits; 16 credits are 2 stills |
How to make a still on this page
- 1
Write a scene with no person
The lines below are original. They do not ask for a 2x video, a 4× video, or a 1080p, 4K, or 8K file.
- 2
Leave the model on Nano Banana 2.1
The picture is 1K and costs 8 credits. 16 credits are 2 stills. This is not the LoRA, not ComfyUI, and not an upscaled video.
- 3
Open the page that matches the other job
The general LTX-2.5 guide is /ltx-2-5, Veo 3.1 Fast, 60 credits. UniFab is /unifab-video-upscaler-ai. The other upscaler notes are /relayvsr and /replicate-nightmareai.
Prompts for the 1K still
Original lines for Nano Banana 2.1. They are not the LoRA's reference video and not a request to upscale a file.
Bar chart
A black glass chart of white horizontal and vertical bars on a dark table, no hands and no brand names.
Star target
A black-and-white radial star target centered on black cloth, soft side light, no camera body and no text.
Empty lightbox
An empty white lightbox on a wooden table, warm lamp, no people and no logos.
Blank frames
A short stack of blank clear film frames on a gray table, no sprocket printing and no hands.
FAQ
LTX 2.5 Upscale questions
The card, the docs, and the API stay side by side. This form's price is 8 credits.
Does this form run the LTX 2.5 Upscale LoRA?
No. The card describes that LoRA. This form is Nano Banana 2.1 at 1K for 8 credits. It does not download weights or return an upscaled video.
Is it 2x, or 2× and 4×?
The card says 2x only and lists one file. The docs.ltx.io usage guide says 2× or 4×. The adapters page says the LTX-2.5 adapter is currently 2× only, and that LTX-2.3 has 2× and 4×. This page does not pick one.
Is LoRA strength a fidelity knob?
The card says strength 1.0 is the intended default, and that fewer steps and lower guidance stay closer to the reference. The usage guide says fidelity is tunable through LoRA strength, guidance, and step count. Both sentences stay.
Is the $0.08 API the same LoRA?
The API page lists 1080p at $0.08, 4K at $0.30, and 8K at $1.20 per second of input, with a short side of at least 480. It does not say the endpoint is this LoRA. The latent file and the 1024×576 Refine Details tile are different lines. The LoRA draft on the card is about 280p.
Is this /ltx-2-5 or UniFab?
No. /ltx-2-5 is the general LTX-2.5 guide and runs Veo 3.1 Fast for 60 credits. /unifab-video-upscaler-ai is UniFab, with Equinox, Vellum, Kairo, and Titanus, not this LoRA. /relayvsr and /replicate-nightmareai are separate upscaler notes.
Is Nano AI part of Lightricks?
No. Nano AI is not affiliated with Lightricks. The name only identifies the LoRA. No logo is shown.
More from Nano AI
Related pages
Nano AI is not affiliated with Lightricks. The name is used only to identify the LoRA. No Lightricks logo is shown. Sources were read on 2026-11-11.
