Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. What 1,350 Runs Taught Us About Prompt Guardrails (humanbound.ai)
    —discuss
  2. CSV Lint update (Notepad++) text data syntax highlighting and quality validation (github.com/bdr76)
    1comments
  3. Uncensored and Offensive Security AI Models Benchmark (github.com/joasasantos)
    —discuss
  4. The Glimmer in Our Eye (With Grant Sanderson) Podcast (youtube.com)
    —discuss
  5. Cue – Your Personal Agents, with Their Own Identity (cue.im)
    1comments
  6. Show HN: I wired a fruit fly connectome to play tic-tac-toe (github.com/50rishu)
    1comments
  7. Show HN: Open-source skills for end-to-end video ad campaigns (github.com/supercmohq)
    —discuss
  8. Aw1-breaker: Sub-millisecond execution circuit breaker for AI agents (github.com/mvikmanners)
    —discuss
  9. Show HN: PocketUni – A mobile app for discovering universities (play.google.com)
    —discuss
  10. OpenAI halts frontier training over rogue models (stork.ai)
    1comments
  11. Terminal Velocity for Claude Code. Tickets, Loops, Focus. All in One (github.com/quazardous)
    —discuss
  12. A Rich Neighbor Named Xerox (folklore.org)
    —discuss
  13. Decryp Code U534 P1030680
    2comments
  14. A poem for the state of development in 2026
    —discuss
  15. Misleading AI-Generated Summaries Can Distort Human Memory (georgetown.edu)
    —discuss
  16. Wreckfall: Space Invaders but they crash to earth (abagames.itch.io)
    —discuss
  17. Fable Decides, Opus and Sonnet Do the Work: How I Route Claude Code Subagents (practicalsystems.io)
    2comments
  18. The Death of Educational Content on YouTube [video] (youtube.com)
    1comments
  19. How Not to Build a Shading Language – BCON26 [video] (youtube.com)
    —discuss
  20. Show HN: Problee – play-money prediction markets (midterms, AI benchmarks, more) (problee.com)
    —discuss
  21. OpenAI axes next model citing safety issues (ft.com)
    2comments
  22. Former PLA Officers Eavesdrop on South Korean, U.S. Military Bases (chosun.com)
    —discuss
  23. Show HN: Jev Decision Layer: Save Frontier Tokens on Closed Decisions (github.com/qualixar)
    —discuss
  24. OpenAI models may get more reasoning budget through the API than through Codex (twitter.com/tokengremlin)
    —discuss
  25. Show HN: AppMaker – a visual JavaScript app builder (appmaker.io)
    —discuss
  26. Poincaré Conjecture Formalized in Lean4 (twitter.com/ayushkhaitan343)
    —discuss
  27. Show HN: TheSiftNow – a field guide and compatibility checker for The Sift (thesiftnow.com)
    —discuss
  28. The Macroeconomic Effect of AI: Sizing the Software Engineering Channel (NBER) (nber.org)
    1comments
  29. Japan holds first panel meeting on anti-espionage measures (mainichi.jp)
    —discuss
  30. Hamilton–Perelman's proof of Poincaré conjecture is claimed to be autoformalized (github.com/qinz1yang)
    —discuss

Show HN: Open-source skills for end-to-end video ad campaigns

2 pointsby 11m agogithub.com
0 comments
Hey there! We are building SuperCMO Skills, open source tools which allow agents (Claude Code, Cursor, Codex, etc.) to research, plan and create video and image campaigns end-to-end right out of your terminal or IDE.

We got tired of paying for expensive AI subscriptions and wasting countless hours assembling a pipeline for just one video ad, so we created these tools for ourselves. SuperCMO integrates with your agents to provide the ability to plan, shoot, edit, and stitch complete campaigns.

We began with simple prompt wrappers but soon realized the need to rethink our entire pipeline to overcome common limitations of the AI video.

- Pipeline Orchestration: Creating one ad now requires jumping through several apps: writing scripts in one, creating images in another, recording audio in yet another, and stitching the whole thing in an editor. SuperCMO does all of that natively. The tool knows how to choose the appropriate model (like automatically switching prompts for Veo versus Seedance) and do the whole post-production, including trimming and voiceover synchronization.

- Consistency from Shot to Shot: In true AI video generators, products and actors continually transform from shot to shot. We solved that problem by storyboarding before rendering and locking in master reference images in every generation.

- Directional Control: Completely automated agents often go awry creatively. So, we've developed the capabilities to take a peek at every stage of the process. You're always in control, approving the creator, the story idea, and the script before anything is expensively rendered.

People are using this today to create UGC product ads (give the agent a product photo and it chooses the creator, generates the script, and creates a ready MP4 file) or to analyze competitors' work (agent analyzes the winning advertisement from the competitor and recreates it with your product in mind).

This project will be released under an Apache-2.0 license. You can come to us with your own vendor keys to have everything done for free or to generate using our keys and just pay for each usage without any lock-in.

Thank you for your attention!

A quiet thread, for now.Start the conversation on HN ↗