Generating four variants adds real value because one can be excellent, but the quality spread means users still have to inspect and compare all outputs to find the best one.
What was measured
Variant Usefulness
Multiple output variants add real value rather than just multiplying low-quality results.
context, not decisivetransformation
Multiple variants can help pick a better result, but they do not by themselves prove the tool adds better sound effects. (3 of 3 judges)
What was given, what came back
Test input: Bird/Nature Video · video · group: auto-sfx-video-generation
Input — what we sent
Input not captured
This run recorded no prompt or input file for the test, so we cannot show you what produced the result below. Capture gaps are tracked, not hidden.
A video featuring a bird in frame, used to test whether generated chirping or ambient sounds sync to the specific animal and movement shown on screen.
Why this input is hard
- · sync accuracy to a visible subject
- · nature/animal sound matching
- · avoiding generic ambience that ignores timing
- · scene-specific audio generation
Output — unretouched
Also checked on this input — same tool, 4 other criteria
Audio Quality✓ WorkedIts background music can sound consistently good across variants, giving the outputs a polished and listenable audio bed.Dialogue & Music Balance✓ WorkedGenerated effects can stay out of the way of an existing music bed, so the music remains intact instead of being overpowered by the SFX layer.Layering Consistency✓ WorkedThe tool can stack generated background music with sound effects smoothly enough that the layers do not conflict.Sync Accuracy✓ WorkedIt can achieve near-perfect timing on natural footage; one variant was reported as so accurately synced that the sound effects seemed like part of the video.
Provenance
- Observation
- 9e609120-c682-42f3-8f2c-210b3f02e0db
- Evidence run
- 90451942-9055-4e8c-85e6-17c773437ed7
- Study
- Automatically Add Relevant Sound Effects to Videos
- Research task
- 86b84w3x5
- Tested at
- not recorded
- Source
- first-party
- Evidence state
- verified
- Proof shown
- input + output shown
- Cost / latency
- not captured
- Repeat run
- not captured
- Tester
- not captured
The last three rows are honest blanks, not placeholders — our capture has no field for them yet.
Query this
get_evidence({
tool: "mirelo",
scenario: "auto-sfx-video-generation"
})MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Same input, same check — 2 other tools
measured on Variant Usefulness
Descript◐ MixedWhen the bird species is ambiguous, the tool can generate four distinct chirp variants (generic bird chirping, birds chirping, tropical birds, morning birds), which adds some choice, but none of the variants fixes the temporal mismatch.Kling AI◐ MixedThe fourth variant can be only marginally better than the others, so the extra variants provide limited benefit.
This evidence is published in
Real inputs and real outputs, no retouching · every cell queryable via API & MCP · aidemos.com