HaoCut frequently asked questions

HaoCut is an AI app for one-click talking-head cleanup, spoken-video search and highlight discovery. It is also your spoken-video asset library. HaoCut transcribes and vectorizes local media so you can search, select and insert clips faster, then continue editing in DaVinci Resolve, Premiere Pro or another professional editor.

HaoCut is for talking-head and course creators, podcast and interview teams, documentary editors, content teams that reuse archived footage, and automation users who want Codex or other agents to work with local video. The longer the footage and the larger the library, the more useful content search and rough-cut preparation become.

No, and that is not its goal. HaoCut handles asset indexing, content search, transcript review, removal suggestions, highlight filtering and rough-cut handoff before footage enters a professional editor. Color grading, mixing, effects, multi-track finishing and final delivery still belong in DaVinci Resolve, Premiere Pro or Final Cut Pro.

Asset indexing, the vector database, local transcription, subtitle search, review marks and local exports run primarily on your computer, so source videos do not need to be uploaded by default. Downloading the app and models, signing in, and any online AI feature you explicitly choose still require a network connection.

The current public download focuses on macOS, with Apple silicon and at least 8GB RAM recommended. Windows is being developed and validated. Check the download page for the latest available platforms and installers.

After you select a local folder as a workspace, HaoCut scans the media, generates subtitles, and uses multilingual-e5-small to vectorize the subtitle text for semantic search across files. Speaker, emotion, sound-event and highlight labels are separate clip-level filters. The current product does not include visual understanding or face recognition.

The indexer recognizes MP4, MOV, MKV, AVI, WebM, M4V, FLV, WMV, MPG and MPEG video, plus MP3, WAV, M4A, AAC, FLAC, OGG, Opus, AIFF and WMA audio. Actual decoding also depends on the codecs inside each file and FFmpeg support.

HaoCut can search all subtitle text in the vector space, then filter by speaker, emotion, sound event and highlight. Each result links to an exact time in the source media so you can play, save or insert it immediately.

Yes. HaoCut builds a local vector index from subtitle text across files in the workspace. Describe the content you want in natural language to find the right clip as quickly as possible, without opening every video or relying only on filenames.

HaoCut uses multilingual-e5-small to convert transcribed subtitle text into vectors and stores them locally for semantic search across files. The database contains subtitle text, vectors and matching time information; it does not copy source video or generate visual, keyframe or face vectors.

Speaker identification can reach 95% accuracy and can save speaker voiceprints for person-based filtering. Emotion analysis supports 4 emotion categories, and audio analysis supports 527 event categories. Results vary with audio quality, noise, overlapping speech and model configuration.

A HaoCut workspace index records the source path. If a source file is missing, its historical index can remain as a record, but playback, reanalysis and export require access to the source. Rescan the workspace to update file status.

Local transcription runs ASR, voice-activity detection and timestamp alignment on your computer, without requiring source media to be uploaded. The selected model must be downloaded on first use. Speed and resource use depend on model size, media duration and computer performance.

HaoCut uses the same local transcription engine as the haoone professional AI subtitle app, with Chinese and English recognition plus word-level audio alignment. Transcription models can be shared with haoone, avoiding duplicate downloads and saving disk space.

The homepage currently lists reference results of 96% Chinese recognition accuracy, 97% English recognition accuracy and 98% word-level audio alignment. These figures come from specific test conditions; actual results depend on the model, recording quality, accent, background noise and specialist vocabulary.

In the reference setup, a Mac mini M4 with 16GB RAM transcribes a 3-minute video in under 20 seconds. Actual speed varies with video codec, model size, audio quality, device performance and concurrent tasks.

Yes. HaoCut can add multiple media files for transcription and show their progress in the task queue. Transcription runs in the background without blocking search, review or other operations.

HaoCut detects repeated takes, filler words and pauses in one click, then uses its in-house repetition-detection algorithm to create removal marks. You can set how much gap to retain and review every mark before export. Source media is never modified directly.

No. AI and rule-based results appear as visible, reviewable removal marks, while the source remains unchanged. HaoCut creates a new rough-cut video or timeline only after you confirm and export, so the result can be restored, adjusted or regenerated.

HaoCut can filter transcript-based highlights and review them in a unified list and timeline. You can define custom highlight rules or instantly save any sentence as a highlight for filtering, insertion or export.

Yes. You can search transcript hits, jump to their timecodes, assign or change speakers, and add removal or highlight marks. Retranscribing replaces the existing transcript and may clear clip marks that depend on its previous time boundaries.

Insert view gives HaoCut a seamless workflow with DaVinci Resolve and Premiere Pro. Search and jump to a sentence, then insert the clip into professional editing software with one click. It is powerful while remaining simple to use.

Since DaVinci Resolve 19, support for third-party plugin interfaces is limited in the free edition, so DaVinci Resolve Studio is recommended for the complete HaoCut plugin workflow. If the plugin interface is unavailable, you can still try HaoCut exports such as DRT, FCPXML or other interchange formats.

Current formats include OpenTimelineIO (.otio), Final Cut Pro XML (.fcpxml), Premiere Pro XML (.xml), DaVinci Resolve timeline (.drt) and a Jianying 5.9 draft package (.zip). You can export highlights or the clips retained after removal marks are excluded.

HaoCut exports SRT, WebVTT, ASS and plain text. You can select all subtitles or a specific speaker and choose whether speaker names appear in the exported text.

You can batch-add clips to DaVinci Resolve or Premiere Pro, export clips after excluding removal marks, export highlights, or combine selected clips into one new video. HaoCut can also generate a prompt for Codex to help decide which clips to use.

Yes. The current free plan includes local video transcription and subtitle editing, basic video-content search, and subtitle and edit-mark exports. Feature boundaries may change by version, so check the pricing page and in-app notices.

The current early-bird price is ¥99, reduced from ¥199. One purchase activates one macOS or Windows computer and permanently unlocks the included local features. The pricing page is the final source for price and entitlement details.

Local transcription, local indexing and local editing do not consume online AI credits. Credits from your HaoAI account may be used only when you explicitly run an online model for AI correction, highlight decisions or another cloud capability.

The current pricing page states a 7-day refund policy. Eligibility, device deactivation and the treatment of consumed online AI credits are subject to the terms shown at purchase and confirmation from customer support.

Yes. Email your company billing name and the price of the purchased software to minghe36@qq.com.

HaoCut can organize candidate clips and generate a prompt so Codex can help decide which clips to use based on the content, context and goal. Fully automated search, timeline writing and Agent client/skill capabilities remain subject to formal release notes.

Planned capabilities include read-only subtitle search, reading transcripts with surrounding context, retrieving source files and timecodes, explaining speaker, emotion, sound-event and highlight labels, and generating reviewable edit plans. File writes, timeline insertion and export should be executed through an authorized client.

The product design principle is that an Agent must not silently damage source media. Candidate clips, removal suggestions and edit plans should enter a reviewable state first; high-risk actions require confirmation and retain the input conditions, matched clips, task ID and export result.

The Codex subtitle-search plugin is currently planned, with no public release date that should be promised here. Check the HaoCut changelog, download page and formal release notes for availability.