Seamlessly Consistent, Photorealistic AI Video with Flux, ComfyUI & Gen3 by Kim Jonghun

From 2D stills to shorts, lip-sync, deepfake and voice cloning: Kim Jonghun (Hundred AI) teaches the full production pipeline behind Korea’s first full-AI video commercial — holding one person’s identity consistent across long photorealistic clips and making them speak with high-sync lip movement.

🎯 What you’ll learn

• Keep one AI person’s identity 100% consistent across long videos
• Flux + ComfyUI professional image workflows (Inpainting, ControlNet, IP-Adapter, AnimateDiff, virtual try-on)
• Text-to-video and image-to-video with Gen3, Kling, Hailuo, Pika, Hunyuan
• Lip-sync and face tools: LivePortrait, FaceFusion, Topaz AI upscaling
• Voice cloning and training: GPT-SoVITS, ElevenLabs, Kohya LoRA, DeepFaceLab
• Finish two portfolio pieces: Fake-TED Talk and Fake-Commercial

✅ Requirements

• Skills: No AI background needed — the course states anyone can follow
• Tools: Stable Diffusion WebUI / ComfyUI setup, Photoshop; accounts for Gen3, Kling, ElevenLabs, Replicate as used
• Hardware: GPU-equipped PC for local SD/ComfyUI work (course covers required gear in CH02-01)

📝 Description

The course is fully dubbed into English: all 68 lessons ship with English voiceover plus English subtitles, so no Korean is needed to follow along.
The publisher’s program spans short-form production to deep learning: 2D images, talking-head video, deepfake and voice training, all aimed at one signature skill — keeping a single character consistent for 30+ seconds of photorealistic footage with synced lip motion. Bonus materials per the publisher: prompts used in lectures, example file lists and links, IDM-VTON/V2V ComfyUI workflows, and deep-learning datasets with labeling examples. Lecture benefits: learn from the producer of Korea’s first full-AI video ad, with immediately applicable production techniques.

Discrepancies, stated honestly: the publisher markets “7 major topics, 60+ clips (about 19 hours)” while the delivered VOD set counts 68 clips at about 20 hours 5 minutes (verified at import against VOD runtimes); the source pack ships no subtitle files (dub Korean subs are ASR-derived); one lesson ships in two near-identical cuts (CH02-13 plain and MUX, ~357 vs ~363 MB) and both are carried through as delivered. The GBC thread title writes “Gen-3” with a hyphen; the poster and catalog use “Gen3”.

🧑‍🎓 Who this course is for

• Creators who want consistent photorealistic AI humans for ads, shorts and virtual characters
• Marketers and filmmakers curious about full-AI commercial pipelines
• Anyone starting from zero with Stable Diffusion / ComfyUI / video AI

🧑‍🏫 About the Author

Kim Jonghun (김종헌) is an AI art director running Hundred AI (헌드레드 AI), credited with producing Korea’s first full-AI video advertisement. He teaches the same Flux → ComfyUI → Gen3 pipeline he uses in client work, from stills through talking video to deepfake and voice cloning.

🏁 Final Result

• Two portfolio videos — a Fake-TED Talk (40s western man lecturing) and a Fake-Commercial (20s coffee CF) — plus reusable Flux/ComfyUI/Gen3 workflows for unlimited talking-AI-human content.

📢 Guys, this is a Group Buy campaign, check the GBC forum thread to get access

You can find the course preview in the download section below.