Spaces:
Sleeping
Sleeping
Commit History
-
Add Beam serverless-GPU captioner (Qwen2.5-VL) with batching
BLIP-base is a 0.25B COCO captioner with no OCR, so on tutorial screenshots it produces generic text ("a computer screen with a website on it") that adds nothing to a step-by-step guide. Serve Qwen2.5-VL-7B-Instruct on a Beam serverless GPU instead; it reads on-screen text and understands UI elements. - beam_app.py: batched endpoint, scale-to-zero, weights cached on a Volume. torch is pinned to 2.7.1 deliberately: Beam hosts run a CUDA 12.9 driver and an unpinned torch resolves to a cu13 wheel, which fails as a silent torch.cuda.is_available() == False and a bare HTTP 500 at request time. - vision.py: Beam backend plus caption_batch(), chunked to keep request bodies bounded. Falls back to the HF Inference API and then local BLIP, so an unreachable or scaled-to-zero endpoint degrades instead of breaking. - guide.py: caption the frame pool in one batched call rather than N calls. - .beamignore: exclude work/ (32MB of frames) and .env from image uploads. Captioning is opt-in: …2d82711
-
Surface Generate errors in the guide panel; raise LLM max_tokens to 4096
Generate failures (missing token, LLM/provider error, no steps) now render in the guide panel itself, not only the easy-to-miss bottom status line — the likely reason the deployed Generate looked broken (the warning was off-screen). Also raise the default output cap so longer guides aren't truncated. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>8b53847
-
Auto-extract: generate the LLM step outline first, anchor frames to step times
Auto-extract now runs guide generation (after transcription) when no draft exists, then extracts frames AT the LLM step timestamps — the same suggestions the per-step weighted selection relies on. Falls back to narration-gated scenes when there is no token/draft. The generated transcript + outline are surfaced in the UI (reused, not recomputed). Smoke test reordered to generate before extracting. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>0e7c368 -
Make auto-frame extraction narration-accurate
- Gate scene-detected frames to the spoken time range (drops screen-recorder intro/idle screens) when a transcript is available. - Weight transcript/LLM step timestamps far more in per-step frame selection: tighter window, exclude out-of-speech frames, and extract a fresh frame at the exact step time when no close frame exists. - Thread the transcript into auto-extract + build; hint users to transcribe first. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>6e3fc02 -
DocuMaker: video to step-by-step DOCX guide (Whisper + HF LLM + BLIP)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>85b485a