after a full research, i found 6 viral channels making full automation youtube videos with ONLY code for $0 per video and 0 manual work 2 week old channels and they earned between $105k and $210k from ads in their first month lol i promised a full free A-Z guide so here it is all of them started this month, combined they already did 21M longform views so, what do you need for this? 1. a claude subscription to access claude code (pay once a month) 2. gpt image 2.5, uses 1 AI image generation per video for the thumbnails, costs cents literally i turned it all into 1 prompt that will make your claude code do the entire thing for you basically when you have claude code open, give it "Bypass Permissions" access, and send it the prompt below and everything will be ready for you to see your first video made infront of you with 0 manual work PROMPT: <inputs> Ask me one question: what-if videos (like Ifverse, IFOLOGY, Ifinite) or discipline films (like Day One Films, Monk Mode Films, Last Soul). Then don't stop until the video is done. </inputs> <research> Channels: youtube.com/@IfverseTales, youtube.com/@IFOLOGYFILMS, youtube.com/@Ifinite-c5h, youtube.com/@dayonefilmsoffi…, youtube.com/@MonkModeFilms, youtube.com/@lastsoulorigin. Study only the ones in my format. With yt-dlp pull every video's views, length, title, upload date and thumbnail. Score each video as views divided by its channel's median. Download the top 3 videos of each channel at 480p with auto subtitles and pull one frame every 2 seconds. Write style.md from what you see and hear: the character, colours, lighting, cut rhythm, camera moves, on-screen text and captions, the story arc with timings per section, the narration, the music, and what the winning thumbnails and titles do that the flops don't. </research> <build> 1. My own character in their style, not a copy: one HTML file, one 1920x1080 canvas, no libraries. A rig of joint angles (hips, torso, head, arms, legs) with every pose their videos use, a walk cycle, and the face on its own layer (blinking eyes that look around, brows, mouths). 2. The script: the topic that beat their channels' normal the most, a new story on their arc, length and pacing. Narration line by line with what we see next to each line. 3. The voice, free: kokoro-onnx locally with the voice closest to their narrator. Exact word timings from faster-whisper. 4. The film: every frame a pure function of time (window.seek(t) draws frame t, no timers). One scene per line, cut on the line's first word. Sets made of flat shapes in 3 depth layers moving at different speeds so camera moves feel 3D. Everything else exactly as style.md says. 5. The render: Playwright seeks every frame at 60fps and encodes the canvas in the page with WebCodecs H.264, 4 subframes averaged for motion blur. ffmpeg mixes the voice, a royalty-free music bed ducked under it the way theirs is, one sound effect per cut, loudnorm to -14 LUFS. 6. The packaging: 3 GPT Image 2.5 thumbnail prompts at 1536x1024 and 5 titles, both following their winners. Put my thumbnail ideas next to their top 6 at 320px wide and keep only what holds up. 7. The machine: wrap it all so a new script goes in and a finished video comes out. </build> <gotchas> Show me a contact sheet of every scene next to their frames before rendering, then fix what doesn't match. The character stays on model in every scene: same colours and scale, only the pose, face and camera change. Never reuse their characters, names or music. If a step fails, fix it and keep going. </gotchas> <start> Ask me the one question, then do the research, build it and render it. Hand me final.mp4, the thumbnail prompts, the titles and the description. </start>