Image · Ideogram 4.5
YouTube thumbnail, 16:9
YouTube thumbnail, 16:9. The man in the first reference photo, keeping his face, short black swept-up hair and skin tone exactly, on the right, chest-up in his black t-shirt, looking at the camera with an excited, surprised expression, crisp rim light. Bold title text "BUILD APPS WITH AI" on the left in three lines, in a heavy condensed sans-serif with a thick dark outline, yellow on deep purple. Behind and beside the title, the 3D scene from the second reference: the dark laptop and phone on raked zen sand with rocks, shown large, cropped to the scene only with no browser window, sidebar or interface. Professional, polished look, warm tan to deep purple graphic background. High contrast, saturated, sharp, readable at small size, bottom-right corner left clear. Reference (Building web apps with AI from inspiration to distribution.md): # Business Update: $100K MRR Milestone - Reached $100K MRR, but revenue mix has shifted dramatically - Aura now represents only ~20% of revenue (down from majority) - Strategy is fundamentally different from before # Finding Inspiration and Product Ideas - Start with what you produce or use every day: identify what’s missing in your workflow - Tools used daily (Codex, Claude Code, Cursor, Gemini, Figma, VS Code) are the best source of ideas - Avoid directly overlapping with big players; find your angle instead - Revisit old products: ask “how would I build this today?” - Shape (icon/illustration library) made ~$150K lifetime, built by hand; now fully reproducible with AI - Angle mockup tool: each mockup used to take 1 day; now recreated entirely with Opus 5.5 + Codex # Building the Tool: Prompting Strategy - One-shot prompt gives an impressive baseline, but is not the end goal - Anyone can replicate a one-shot; the edge is in breaking it into parts - Frame the problem thoroughly before building: - Describe desired outputs (YouTube covers, Twitter posts, motion design, mockups) - Name the tech stack: React + Vite (safer and more open than Next.js), Netlify for deployment - Mention future needs early: Supabase (database, real-time collaboration, teams), Stripe (Managed Payments handles taxes automatically) - Give quality references per feature: Figma for design tooling, Spline for 3D, Linear for landing page - “Self-verify until it’s perfect”: AI builds, references the target, and iterates in a feedback loop - Use meeting notes (e.g. from Granola) as raw context to feed directly into the prompt - Push the AI to do everything autonomously; only intervene when it’s truly blocked # Integrating APIs - Adding an AI layer elevates a tool from static to generative - Key APIs worth knowing: - Hugging Face: image, video, 3D generation - Claude API: best for code generation, illustrations, SVG - OpenAI API: image, voice, sound effects, music - ElevenLabs: audio, voiceover, music generation - HeyGen: avatar and UGC video generation - Prompting pattern for API setup: name the API, tell the AI to log in, create and store the key securely (env variable or Supabase/Firebase), then use it - Codex handles API keys and auth flows better than Claude Code # Craftsmanship in the AI Age - Transition: everything that happened in Figma + VS Code now happens via prompt - Designer’s role shifts to QA and judgment, not manual execution - Ask AI to score output out of 10, find mistakes, and return screenshots of issues - Review those screenshots as a manager would: approve fixes, flag problems - “If you care, the AI cares. If you don’t care, the AI doesn’t care.” - Discipline and taste must be passed to the AI through how you p Variation: Use a clean premium tech look: a smooth tan-to-royal-purple gradient with soft light streaks and subtle lens flare, and the zen sand scene seen from a top-down three-quarter view with the rocks as depth layers. Give the man a crisp cool cyan rim light for contrast against the warm tones, with the title in glossy yellow with a subtle drop shadow.
Meng To@shadownessguy0097- Model
- Ideogram 4.5
- Made by
- Meng To
- Shared
- Likes
- 0
Use its prompt, turn it into a video or a 3D model, or find more like it in Fil. Fil is invite-only for now: without an invitation, Open in Fil shows the waitlist.
