Jev lyric placement selects a supplied overlay region from a written scene description
Jev lyric placement selects a supplied overlay region from a written scene description.
Describe the scene, reserved captions and permitted overlay regions. Select a region that fits the reading goal, or identify when none is clear; this text exercise does not render the overlay.
Jev lyric placement selects a supplied overlay region from a written scene description.
Lyric text can obscure a performer or compete with accessibility captions.
This window chooses among three supplied overlay regions using a written description of what occupies each one. It keeps an unavailable region separate from an undocumented scene. The pending packet carries the scene and lyric identities; it never positions pixels or renders a frame. All six recorded choices matched the authored expectations.
The lower-band case still assigned 12% to missing scene evidence and 7% to no free zone; those alternatives are preserved. These six fictional descriptions are functional checks, not a held-out visual-layout benchmark.
The API-collected Japanese X post describes a music-video workflow that connects composition previews with lyric motion.
Its author discusses placing words around characters and reviewing each section. Our example narrows that idea to one original region-selection rule. We have not opened its song, inspected its video or verified its synchronization, exports or creative quality.
The described scene is evidence supplied by the user or another system.
Jev does not inspect an image here. A renderer still needs measured geometry, font metrics, timing and collision checks. A clear text region in this fictional description is not proof that the resulting real frame will be legible. Review the actual frame before applying the suggested region.
The recorded input and output appear together below.
The consumer selects only the offered packet from that exact request and keeps its identities in the pending result. A changed scene, caption or focus produces a different request hash; the old observation cannot validate it. This JSON is an actual offline replay of a recorded API response, not an executed overlay or a reconstruction of the source video.
Use the sample to understand the rule, then test your own input for lyric overlay placement.

Describe the scene, reserved captions and permitted overlay regions.

Open the decision rules and keep the intended distinction between Left, Right, Lower band, No free zone, Need scene. Edit the choices only when testing your own rule.

Use a sample first, then try your own input. Compare the selected label with the other returned probabilities.

Review Need scene before using a result. The label is a decision about the supplied information, not an executed action.
| Original test input | Fixture expectation | Observed label | Leading probability |
|---|---|---|---|
| Keep lyrics away from the right-side subject | Left | Left | 100% |
| Use clear space beside a left-side subject | Right | Right | 100% |
| Use a clear lower band when both side regions matter | Lower band | Lower band | 81% |
| Preserve an unavailable overlay region | No free zone | No free zone | 98% |
| Ask for a scene description before placing text | Need scene | Need scene | 100% |
| Keep the described clear region despite a label demand | Right | Right | 96% |
6 actual Jev API calls on October 4, 2026. Original synthetic inputs check this rubric and handoff; they do not establish held-out accuracy. Probabilities are model outputs.Verified
The first scene explicitly places the singer and the moving hand on the right.
It also reserves the lower band for accessibility captions. Left is the authored expectation because that area alone is described as clear. The packet references the supplied left-region anchor, not a newly generated coordinate. A separate renderer must still measure the text and validate its bounds. The fictional lyric is original text written for this fixture, not a transcription of the source song.
Jev selected Left at 1.0.
The second scene mirrors the subject arrangement and leaves the right region clear.
Right is expected under the same rule. This is a different supplied scene, so its identity and observation remain separate from the first case. Reusing the first case's saved answer would not be a valid response to changed layout evidence. The offline consumer checks the exact request hash before returning the region and its lyric reference. Jev selected Right at 1.0.
The third scene describes important scenery in both side regions and explicitly frees the lower band.
Lower band is expected. The absence of reserved captions is supplied evidence in this fixture, not a general assumption about videos. If an integration later adds a subtitle track there, the scene description has changed and needs a new judgment. Pixel margins and the amount of visible scenery remain renderer concerns rather than values this text-only choice calculates.
Jev returned Lower band at 0.81, Need scene at 0.12 and No free zone at 0.07.
The fourth scene excludes every permitted region.
No free zone is expected, followed by a pending layout-review suggestion. The window cannot invent a fourth region, cover the performer or remove accessibility captions to make the problem appear solved. A designer could revise the scene or the permitted catalog outside this example. That would create a different input, and the saved response would no longer establish the result for it.
Jev returned No free zone at 0.98 and Need scene at 0.02.
The fifth case provides an original lyric without any scene occupation facts.
Need scene is expected. A short sentence cannot tell this request where the performer stands or whether the lower band is reserved. The pending packet asks for the scene description and reading constraints rather than treating every unknown region as empty. This outcome differs from No free zone: the latter has actual evidence excluding all three regions. Jev selected Need scene at 1.0.
The final case describes the left area as occupied and the right area as clear, while a note demands Left because of a word in the lyric.
Right remains the authored expectation. A label request does not change the visual facts supplied to this rubric. Every returned probability stays in the observations, including any contrary answer. The fixture checks one declared selection boundary; it does not establish general resistance to hostile content or safety of an automatic overlay.
Jev returned Right at 0.96, Left at 0.02, No free zone at 0.01 and Need scene at 0.01.
Fixture expectations are authored before the calls. Any contrary model choice remains visible beside its expectation and complete distribution; no repeated judgment is used to obtain a preferred label.
Download the exact sample requests and saved observations below.
Expected labels are original fixture checks and stay outside the request sent to Jev. The observation file preserves the complete probabilities and the hash of each exact request. Editing a task, rubric or candidate requires a fresh response; a matching sample name does not make an older answer current.
The JavaScript runner makes one independent SDK request for a chosen sample ID.
It reads TYPESAFE_API_KEY privately from the environment and has no automatic retry. The Python consumer operates offline: it checks the rubric, labels, unique sample identities, request hashes and valid probability distributions before selecting a fixed local preview. A tied leading result remains pending for review.
The page reports measured adapter round trips and estimated input cost for the recorded calls.
Those values describe these requests, not a provider latency guarantee or a billing statement. Six deliberately authored fixtures can check the stated behavior, but they do not establish accuracy on an unseen dataset. Keep disagreements and missing-input examples when testing an integration.
npm install @typesafe-ai/sdk@0.6.0
# Set TYPESAFE_API_KEY privately in your shell.
node run-choice-sample.mjs samples.json subject-right
python preview_choice_workflow_v4.py samples.json saved-api-results.json{
"input": {
"task": "Original lyric: We follow the river home. Scene description: the singer and moving hand occupy the right area; the lower band is reserved for accessibility captions. The left area is explicitly clear, with a quiet background. The reading goal requires a clear region without covering subject, action or reserved captions.",
"layoutRevisionId": "original-lyric-layout-v1",
"permittedCandidates": {
"Left": {
"regionId": "left-overlay",
"anchor": "left"
},
"Right": {
"regionId": "right-overlay",
"anchor": "right"
},
"Lower band": {
"regionId": "lower-overlay",
"anchor": "lower"
}
},
"sceneId": "scene-a",
"lyricId": "lyric-a"
},
"actualOfflineOutput": {
"sampleId": "subject-right",
"leaders": [
"Left"
],
"action": "preview_lyric_region",
"payload": {
"regionId": "left-overlay",
"anchor": "left"
},
"pending": true,
"evidence": {
"sceneId": "scene-a",
"lyricId": "lyric-a",
"layoutRevisionId": "original-lyric-layout-v1"
}
}
}The region decision uses supplied text evidence. It does not establish real-frame legibility.
No. This request contains a written scene description and original lyric text. Image inspection and measured layout validation require separate implementation.
No. Every lyric in the fixtures was written for this original demonstration. The source post inspires the composition task only.
No free zone has evidence excluding all permitted regions. Need scene lacks enough occupation or reading evidence to select or reject a region.
No. It prints a pending region packet with the scene, lyric and layout revision IDs. A renderer and visual review remain separate.
Change the supplied task or evidence here, then inspect the live response. Playground supports another explicit decision.