NextNeural/NINA

Bramka tematu dostaje przyklady, bo sama regula jej nie wystarczyla

Jezyk odpowiedzi idzie za biezacym pytaniem, a bramka przestaje odsylac pytania o nagranie

Film bez sciezki dzwiekowej przestaje wywracac cala analize

Poprawka linii logu: pipeline.py nie importuje os

Nieudany odczyt wideo zostawia slad w logach

NINA dostaje osobowosc: montazystka, ktora mieszka w internecie

Limit wideo 200 -> 500 MB

Wczytane wideo przezywa odejscie sluchacza

Follow-up z kontekstu szybkim modelem, z limitem na sprawdzenia i z logami

Brak sekretu to odmowa, nie przepustka

fetch: add proxy support + cookie/proxy diagnostic logging

Fix diarization hub kwarg, yt-dlp cookies, stronger output formatting

Add pasted-link video ingestion (yt-dlp) + /fetch endpoint

Upgrade VLM to Qwen3-VL-30B-A3B-Instruct

Re-analysis confirm route + adaptive answer formatting

Visual-only answers, coreference + on-topic guard, 15-min cache

Stream real per-stage progress markers during analysis

NINA chat: NEED_VIDEO handshake for one-video-per-chat

NINA: conversational chat endpoint (Phase 2)

NINA: stop identity hallucination, fix diarization shims, drop output header

NINA: VLM-centric synthesis — the model now actually sees the video

NINA: fix diarization (torchaudio AudioMetaData shim) + OCR score parsing

NINA: pin rapidocr-onnxruntime==1.2.3 (newest that supports Python 3.13)

NINA Phase 3b: OCR burned-in captions (RapidOCR) as the last fallback rung

NINA Phase 3a: embedded subtitle fallback (no new deps)

NINA: fix Space build — torch/torchaudio 2.11 matched pair; pyannote token kwarg

NINA Phase 2: vocal isolation + speaker diarization + name attribution

NINA: anti-hallucination ASR + honest low-confidence handling (Phase 1)

NINA: fully parallel Whisper ∥ CLIP, fused at summary

NINA v0.82 — upload compression, natural multilingual answers, robust precise

Fix synthesis IndexError: guard against empty choices in HF stream events

NINA: SSE keepalive for mobile — prevent stuck 'Uploading & analyzing'

NINA v0.91 — 5-part retrieval + accuracy upgrade

v0.90

NINA v0.90 — anti-hallucination pass + DEBUG-gated output

NINA v0.81 — modular refactor + usage-token gate + 8-vCPU queue

DEBUG TRUE app.py

NINA: privacy/cleanup for uploaded media (delete_cache + idle eviction)

NINA v0.80 — performance pass (app.py)

Update README.md

Rename main.py to app.py

Add gitignore

NINA v0.52 - Transcript prompt-based summarization

NINA v0.0 - video transcription

initial commit