Turn one photo into a 3D object you can spin around, sides and back included, entirely on your phone.
On a Galaxy S26, Idol (speridlabs) on LiteRT now makes a splat of 65,536 Gaussians in 158 s with its new default form, or of 262,144 Gaussians in 267 s with the first release's form, which stays in the repo (the first photo after start, on a cool phone). Checked against the author's PyTorch pipeline, renders from four sides agree at a PSNR of 32.3–34.4 dB for the default and 41.4–46.6 dB for the larger form.
MiniCPM-OCR-Decision-2B on a Galaxy S26: 861.6 ms of LiteRT calls to score one text region, on the GPU.
OpenBMB's OCR checker, converted to LiteRT: a score from 0 to 1 for an OCR result.
In the clip, an invoice for an expense report. OCR read "s1.95", the photo says "$1.95": 0.00, Review. One character fixed: 0.92, Accept. Of 6 lines, the 3 misread amounts go to review, at a median 1,255 ms a line.
d1-2B on a Galaxy S26: four answers about one photo, boxes included, in 1,983 ms.
Liquid AI's decision model for text, pictures and video, converted to LiteRT: one forward pass, no tokens generated, typed answers (yes or no, a choice, a number with an interval, a value with its span, boxes) with their probabilities.
In the clip, a photo of a bicycle in snow (CC0) and four questions. One tap: every wheel, 2 boxes (found 1.00 and 0.89); the saddle, one box (confidence 0.94); how many bicycles, 1 (80% between 1 and 1); a car, No (P(yes) 0.08). Answered in 1,983 ms on this phone, 423 tokens on the 512-token graph, S26 GPU through LiteRT.
To run it: the repo has the LiteRT graphs (five also compiled for the S26's NPU) and a Python host for desktops, no PyTorch. The app in the clip is my Kotlin sample: add a photo or paste a message, write your own questions. No internet permission. https://github.com/john-rocky/LiteRT-Models/tree/main/d1-2b
LFM3-Audio-2B on a Galaxy S26: 40.6 tokens/s on the GPU, 1 s of speech in 0.40 s, no network.
Liquid AI's 2B audio model, converted to LiteRT: text or audio in, text, speech or both out. Transcription, synthesis and translation in 18 languages, voice chat in English (their card).
In the clip, the phone is in airplane mode. A spoken question, "What's a quick way to fall asleep?" (a TTS voice), one tap, and the answer is spoken as it is written, the first sound 1.4 s after the tap.
To run it: the .litertlm file works with LiteRT-LM 0.18.1 on Android and Mac, text out from text or speech; speech out takes four more LiteRT graphs from the repo and a loop in your app. The app in the clip is my voicemail sample (Kotlin): an English voicemail, told back in your language as text and speech, with the number to call back. No internet permission. https://github.com/google-ai-edge/litert-samples/tree/main/samples/litert/voicemail_summary
