# HyperFrames + Pocket TTS video upgrade Status: `added_as_safe_video_workflow_note` Skill installed locally: `hyperframes-pocket-tts-video` Official Hermes skill installed locally: `hyperframes` from `official/creative/hyperframes` Public actions: `closed_until_human_approval` ## What just got added We added a new Hermes/Jimsky skill for the Sonic-Forage video lane: > **HyperFrames + Pocket TTS Video Workflow** This gives Afterparty Forge a cleaner way to make and explain videos: - **HyperFrames** turns videos into HTML/CSS/JS compositions that agents can edit like web pages. - **Pocket TTS** gives those videos swappable local voices. - **FFmpeg** muxes the final video/audio and exports platform variants. - **Hermes/Jimsky** can write the composition, generate the voices, build drafts, verify files, and prepare manual-upload packs. The important change: videos become **remixable source code**, not one-off exports. ## Launch explanation Short version for a video or Space: > We just added a video-native layer. HyperFrames lets the agent write videos like web pages — HTML, CSS, timing, animation, and media assets in git. Pocket TTS gives the same pipeline swappable local voices, so the narrator can become Alba, Javert, Cosette, or a different character without rebuilding the visuals. That means Afterparty Forge videos are no longer one-off exports; they become remixable source code. ## What Hermes can do with the new skill Hermes can now help with: 1. Scaffolding HyperFrames-style video projects when the Node toolchain is ready. 2. Writing timed HTML scenes for launch videos, Shorts, explainers, and proof-hub walkthroughs. 3. Turning Sonic-Forage docs, launch copy, or proof cards into a narrated video structure. 4. Generating Pocket TTS narration locally. 5. Changing the narrator voice per scene. 6. Creating multi-speaker scripts: host voice, entity voice, DJ voice, announcer voice, outro voice. 7. Replacing a video voice track without rebuilding the whole visual layer. 8. Exporting Telegram, Discord, X, Shorts, and YouTube-ready variants. 9. Preparing captions/transcripts and manual-upload checklists. 10. Keeping posting, streaming, uploads, dataset-public release, and money actions closed until explicit approval. ## Current local verification HyperFrames / render stack: - `npx hyperframes --version` returned `0.4.45`. - Local Node is `v20.19.2`. - HyperFrames docs recommend Node `>=22`, so full HyperFrames rendering should be treated as **not fully verified** until Node is upgraded or a Node 22 project runtime is used. - FFmpeg is installed and available. Pocket TTS stack: - Pocket TTS project exists locally. - OpenAI-compatible Pocket TTS server was started on `127.0.0.1:49112`. - Health endpoint returned healthy. - Available server voices verified: - `alba` - `azelma` - `cosette` - `eponine` - `fantine` - `javert` - `jean` - `marius` Generated voice proof: - MP3: `docs/assets/audio/hyperframes-pocket-tts/hyperframes_skill_alba.mp3` - OGG/Telegram voice version: `docs/assets/audio/hyperframes-pocket-tts/hyperframes_skill_alba.ogg` - Voice used: `alba` - Approx duration: 18 seconds ## Voice swapping pattern Scene manifest example: ```json [ { "scene": "cold_open", "voice": "alba", "text": "We missed the party, so we built the afterparty." }, { "scene": "proof_stack", "voice": "javert", "text": "The proof hub is live, the repo is public, and the gates are closed." }, { "scene": "builder_invite", "voice": "cosette", "text": "Come test, fork, break, remix, and shape it with us." } ] ``` Each scene gets its own generated voice clip. The clips are placed sequentially on the timeline and muxed into the video. ## Pocket TTS request pattern Verified minimal request: ```python import json, urllib.request payload = json.dumps({ "input": "We missed the party, so we built the afterparty.", "voice": "alba" }).encode() req = urllib.request.Request( "http://127.0.0.1:49112/v1/audio/speech", data=payload, headers={"Content-Type": "application/json"}, ) with urllib.request.urlopen(req, timeout=60) as resp: open("voice.mp3", "wb").write(resp.read()) ``` ## Official Hermes HyperFrames skill alignment The official optional Hermes skill from the docs is now installed as `hyperframes`. Install source: ```bash hermes skills install official/creative/hyperframes ``` Official workflow rules now adopted for this video lane: - **HTML is the source of truth for video.** - Define visual identity before writing composition HTML; avoid generic defaults like `#333`, `#3b82f6`, or `Roboto` unless they are explicitly in the design system. - Build the **hero frame** first, then add GSAP animation. - Use `npx hyperframes lint`, `npx hyperframes validate`, and `npx hyperframes inspect` before render. - Use `npx hyperframes render --quality draft --output draft.mp4` for drafts and high quality only for final local verification. - Multi-scene pieces should use transitions instead of hard jump cuts. - Full render verification still waits for Node `>=22`; the current machine is Node `v20.19.2`. ## HyperFrames composition pattern Minimal shape: ```html

We missed the party. So we built the afterparty.

``` ## Safety gates This upgrade does **not** mean videos auto-publish. Still closed without explicit human approval: - Posting to X/Twitter, Discord, YouTube, Reddit, LinkedIn, etc. - Scheduling or starting an X Space. - Starting livestreams. - Uploading or publishing videos. - Making private Hugging Face datasets public. - Spending money or starting paid GPU/API jobs. - Creating payment links or invoices. - Claiming endorsements, affiliations, revenue, users, or traction. ## Next practical use Best next video increment: 1. Pick one video target: X teaser, 60-second proof hub walkthrough, or YouTube intro. 2. Choose 2–3 Pocket TTS voices. 3. Write a scene manifest. 4. Generate voice clips. 5. Create HyperFrames or PIL/FFmpeg visuals depending on Node 22 readiness. 6. Export a draft MP4. 7. Verify locally. 8. Ask for manual upload/post approval separately.