正在加载视频...
视频加载失败
Your AI UGC looks fake because you're solving the wrong problem. It's not a prompting problem. It's a visual problem. Here's the workflow that's producing native-looking AI B-roll in under 5 minutes🧵:
14,337 次观看 • 5 个月前 •via X (Twitter)
17 条评论

Step 1: Pinterest before prompts Search the shot you need. Find a real photo with the exact lighting and framing. Don't describe the scene — find it.

Step 2: Kill the background blur Blurry backgrounds = camera. Flat backgrounds = phone. If you want UGC that looks native, become the anti-blurry-background person. Every time.

Step 3: Use Nana Banana Pro (not v2) It was trained on iPhone images. It has a better reference bank. Use it to swap the person, product, or setting — but keep the lighting locked.

Step 4: That image is now your base Every other B-roll shot you need? Use this as your character reference. Consistent lighting. Consistent feel. No re-prompting from scratch.

Step 5: Animate with the right model Kling 3.0 Pro → best skin texture Sora 2 Pro → best overall realism Google Veo 3 → skip it Step 6: Write your video prompt like this "Handheld video of woman shopping, shot on an iPhone." No "cinematic movements." No "make it look realistic." Just describe what's happening, present tense.

The people getting the best AI ad creative aren't better prompters. They're better at finding references. Want the full system — prompts, frameworks, weekly ad roasts, and new creative formats every month? Join Ad Creators Lab 👇 🔗

@Camicees Well said To make realistic AI UGC, you need to actually have a creative eye and attention to detail. Otherwise you're cooked

Steal our agencies' prompts, frameworks, and systems to make ads with AI that scale

@Camicees Visuals drive trust, not just prompts

@Camicees what is step 3

@Camicees @grok turn this thread into a wor doc

@Camicees Fire <3

“It’s a visual problem” is the whole game. Find the real shot, kill the cinematic blur, then animate. Everything else is noise.

@Camicees The visual system matters more than another prompt tweak.

@Camicees Which model did you use for this? try

@Camicees The only limit is creativity rn

In the current times the best biggest bottleneck is the voice
