Live Mood Visuals MVP

Mic -> STT -> scene intent via selected LLM provider -> Pexels image -> smooth hero transition.

Execution mode

Changes both scene-analysis backend and transcription backend. Model list below follows the selected mode.

Model / deployment

Audio source

Input level: idle

Pexels access password

Pexels access Not verified

          Prefer server STT (`/ollama/transcribe.php`)
          
            Meaning change threshold
            
              Controls how different the next scene phrase must be before a new image is fetched. Lower value updates more often; higher value is more stable and updates less often.
            
            Audio chunk seconds (4-8)
            
              Defines the speech chunk length sent to transcription each cycle. Shorter chunks react faster but can be noisier. Longer chunks are steadier but slower to update visuals.
            
            Server STT timeout seconds
            
              How long to wait for transcription before giving timeout. Increase this for multilingual or larger Whisper models if you see timeout errors.
            
          Language
          
          Pexels orientation
          
          Pexels size
          
          Pexels color filter
          
          Pexels locale
          
          Pexels candidates (1-10)

Idle. Enter Pexels access password, then press Start.

Current transcript window (1-2 sentences):

Live transcript stream:

Pipeline timings (ms):

LLM verdict (latest):

Pexels search used (latest):

Not running

Waiting for speech...

Press Start and begin talking.