Kian | AI Builder

PitchFrame won 1st place at the HeyGen Hackathon. I built it to turn rough project briefs into presenter-led pitch videos, with HeyGen as the human layer and Hyperframes as the programmable motion layer. Thanks @HeyGenDev for hosting.

Lately Codex GPT-5.5 high/xhigh feels noticeably slower. More reconnect loops, websocket disconnects, and long agent runs stalling mid-response. Feels especially rough on large repos + tool-heavy workflows. Wondering if others are seeing the same lately...

Forget trying to master the 'perfect' massive app. Instead, just start by launching a $5 mothly service that solves one tiny, specific frustration. I'm continuing to build every day. Let's make something lightweight and execute. 🚀 Inspiring advice from Gary Vee: youtu.be/4vIIeCqHYXA?si… #buildinpublic

After PitchFrame, I wanted to build a smaller, sharper tool for creator video workflows. Typeframe turns spoken video into speech-timed, typewriter-style captions frame by frame. Upload, transcribe with Whisper, tune the typography, export an MP4. Planning to launch this week

Links: Live demo page: pitchframe-flax.vercel.app/hackathon Generator app: pitchframe-flax.vercel.app The generator is BYOK right now: bring your own OpenAI, HeyGen, and fal.ai API keys. Hosted runs use object storage for uploaded inputs and rendered outputs. I’m still deciding whether this should stay hosted, become a local Tauri/Electron app, or support both.

After winning 1st place at the HeyGen Hackathon, I made a full walkthrough of PitchFrame. PitchFrame turns a rough project brief into a HeyGen-presented pitch video with Hyperframes motion, Revision Studio, purpose cuts, and a product feedback loop. Still early, but the end-to-end loop works. Next focus: speed, quality, and deeper editing control. Thanks again @HeyGenDev. Full demo: youtu.be/kXv1-7AL-pc

I posted a full walkthrough of PitchFrame after the hackathon. It covers the end-to-end flow from project brief to HeyGen presenter, Hyperframes motion, Revision Studio, purpose cuts, and the product feedback loop. Full walkthrough: x.com/vericontext/st…

I’m planning to update the Vercel URL with a more polished demo and clearer explanations.

Built PitchFrame for the HeyGen Hackathon. It turns a rough project brief into a HeyGen-presented pitch video: script, presenter, Hyperframes motion, QA repair, purpose cuts, and PostHog feedback loop. @HeyGenDev

@HeyGen Huge thanks to the HeyGen team for hosting such a great hackathon. It was a lot of fun building PitchFrame with HeyGen and Hyperframes, and I'm excited to keep exploring what programmable AI video workflows can become.

Catch the recap of our livestream: x.com/HeyGen/status/…

33 teams entered the first @HeyGenDev Hackathon to build scalable AI video workflows, agents, and creative tools with HeyGen The level of execution blew us away. Huge thanks to everyone who built with us Winning projects and demos below ↓

Grand Prize ($3000): Pitchframe One of the strongest entries. A real agentic generate → lint → inspect → repair workflow built around deep HeyGen v3 + HyperFrames integration with production-level execution x.com/vericontext/st…

@HeyGen Thank you for hosting such an amazing hackathon! Grateful PitchFrame was selected as 1st place, and excited by what HeyGen + Hyperframes make possible for programmable AI video workflows 🔥

HeyGen Hackathon: Demo Day x.com/i/broadcasts/1…

Been heads down building. Planning to launch Framlit 2.0 + Typeframe soon. Now the real work begins: making demos people actually remember.

@mdmadeit In the case of Seedance, does giving keyframes for short time segments within a single shot and writing prompts for each segment help improve the video generation quality?

back then it was practice now it’s survival made with gpt 2.0 and seedance 2.0 prompt + storyboards below

The prompt is in the above link.

I also wired up a publish loop: VibeStudio export → Cloudflare R2 media upload → git push → GitHub Pages gallery. Still early, but it's becoming a useful way to compare GPT Image 2 character sheets, storyboard prompts, provider prompts, and final video results without wasting API spend rerunning everything vericontext.github.io/vibestudio/

Built a small OSS studio around a workflow I've been testing: generate a character sheet with GPT Image 2, then use it as a reference in Seedance 2.0 reference-to-video. It handles storyboard drafts, clip generation, preview, and export from one local UI.

@Zyrellix This is a one-shot prompt for text-to-video with seedance 2.0, right?

Created with Seedance 2.0 Prompt: An intense, cinematic, slow-motion action sequence on a damp, overcast city street during dusk. A young Asian man in a black hoodie is in a tight, fast-paced hand-to-hand martial arts fight with a glowing, high-tech holographic or energy-based blue dynamic cyborg. The camera dynamic moves quickly, capturing sharp dodges, blocks, and misses. VFX & Transformation Sequence: The man jumps back into a defensive stance. A close-up reveals his forearm glowing with intricate, incandescent orange and gold circuit patterns, emitting light steam. With a dramatic fist clench, an explosion of golden sparks and metallic nano-particles erupts, rapidly morphing and crawling up his arm to form a high-tech gold and black metallic gauntlet. The camera pans out as the nano-armor swiftly encases his entire body, forming a sleek, studio-quality gold and black exo-suit with a fabric hood. The hero extends his arm, generating a swirling vortex of golden energy that fires a powerful concentrated beam across the street, illuminating the wet asphalt. Lighting & Atmosphere: Moody, desaturated overcast daytime lighting with a soft sunset glow in the background. The scene features highly reflective surfaces—wet ground, glossy cars (Hyundai SUV), and polished metallic armor. The hero’s helmet features a smooth, reflective gold visor that suddenly activates, glowing with sharp, vibrant orange LED eyes. Ultra-detailed, photorealistic, 8k resolution, cinematic composition, Hollywood-style visual effects.

Trains billion-parameter LLMs from scratch on a single GPU github.com/FareedKhan-dev…