Spaces:
Running on Zero
Running on Zero
Commit History
2 days ago
-
fix: each sample's caption says it was made earlier by this app from its example, with that example's inputs and seed, and no longer promises the same result from the same prompt and seed (#20)
Unverified signatureRound 4, minor 8. The samples were made on the operator's Mac (Q8 GGUF transformer); on the Space the same prompt and seed give another take (mode 05 at seed 12: two crewmen on deck there, the ship from above in the sample), yet the captions said "Same prompt and seed as “Storm on deck”" or "same seed", inviting a Space visitor to expect the sample from 360 s of free quota. Every caption now reads "Made earlier by this app from “<example>”, with its <prompt / picture / clip / recording> and seed: ...": what the sample is and where it came from, with no promise of a repeat. The words live in examples/manifest.json. Test (tests/test_ui_sample_captions.py): every shipped caption starts "Made earlier by this app from “" and says neither "Same prompt and seed" nor "same seed".b46dfc2 -
fix: the sample captions name the example each was made from, state the samples' real lengths ("4.7 s, the longest that fits the 5.0 s recording") and say "no upscale" in place of "single pass" (#20)
Unverified signatureRound 4, minor 11. Modes 10 and 11 called 4.7 s "the recording's length" for a 5.0 s recording (examples/media/ocean.wav measures 5.000 s; the run is the longest length on the frame grid that fits it, 113 frames), mode 11's "From the example's recording and prompt" did not say which of its two examples, and "single pass" is builder shorthand. Each caption now names its example in quotes ("Made by this app from “Waves, pictured”, with its recording, prompt and seed: LTX-2.5, no upscale."), the video modes say "no upscale", and the settings of modes 10 and 11 read "4.7 s ..., the longest that fits the 5.0 s recording". The words live in examples/manifest.json, beside the samples they describe. Tests (tests/test_ui_sample_captions.py, over the shipped manifest): every caption names exactly one of its mode's examples, whose seed is the sample's, and carries no "single pass" or "the example's"; every video mode's says "no upscale"; every length a sample's settings state is the file's own …9ab0812
3 days ago
-
fix: every sample on the page is a re-encode of the final real-weights pass, and audio to video's shows an ocean recording pictured as waves at sunset, not a speaking face
Unverified signatureSix samples change, all through Git LFS (the index holds three-line pointers): - image_to_video.mp4, text_to_audio_video.mp4, image_to_audio_video.mp4 and image_text_to_audio_video.mp4 were made on 75d0c53, one commit before the float64 RoPE fix; they are now F02, F05, F06 and F07 of the final pass on e59944e; - text_to_audio.m4a, also from 75d0c53, is now F08, run on ba2d582 after the early memory reading had refused it on e59944e; - audio_to_video.mp4 is now F11b_ocean: the example "Waves, pictured" (the ocean recording and a beach prompt, seed 23), waves rolling onto a beach at sunset with gulls, the recording itself as its sound track. The other five samples were already byte-identical to the final pass's re-encodes (their runs are identical to the earlier ones). examples/ is now byte-identical to the lead's set (sha256 over all 19 files).54787af -
fix: audio to video's examples are an ocean recording pictured and a portrait over the sea in which nobody speaks, its two speech recordings leave the repo, and a test refuses an example of that mode that asks for speech
Unverified signatureThe round-3 UI review (R3) found mode 11's showcase made of speaking faces, the one thing the page itself says the mode cannot do: the sample was the captain's portrait set to his spoken line, and both examples were speech ("A portrait set to a recording" with captain_line.wav and a shouting prompt; "A voice, a studio" with studio_welcome.wav and a prompt that speaks to the camera). Measured on real weights, the mouth follows the words at +0.845 in image + text to video with sound and at -0.379 in audio to video. examples/ now matches the lead's set built from the final real-weights pass (assets/ui_examples_v3), text files first: - manifest.json: mode 11's examples are "Waves, pictured" (ocean.wav and a beach prompt, 3:2, seed 23) and "A portrait over the sea" (ocean.wav, captain.jpg and a prompt in which his lips stay closed, 480 x 704, seed 24); its sample is the first one's output, captioned "From the example's recording and prompt, same seed". The settings lines of the 05 …b7f0260 -
feat: the page ships an example and a sample output for every one of the eleven modes, from examples/ through Git LFS
Unverified signatureexamples/ is the loader's default directory. The samples are re-encodes of this app's own verified real-weights outputs, the inputs are the project's test assets; examples/PROVENANCE.md says where each file comes from. Every media file is committed as a Git LFS pointer (git lfs ls-files lists all 19; no binary in the numstat). An export with `git archive` in a clone that holds the LFS objects carries the media themselves; GIT_LFS_SKIP_SMUDGE=1 gives the pointers. tests/test_examples_shipped.py: the manifest names all eleven modes, each with an example and a sample; every path it names is in examples/ and every media file is named; .gitattributes keeps the media in LFS and the manifest plain. Reading the first bytes of one sample decides which checkout this is: with the media fetched the loader returns every example and sample and the default mode's sample plays on the monitor; with pointers it skips each entry that needs one, with its log line, keeps the text-only examples, and the pa…e59944e