Dam

AI videos are getting too good we’re cooked. here's the full workflow with prompts, step-by-step: 1. find a real UGC video on TikTok as a reference for style, emotions, and movement (style reference not a copycat video with different face) 2. create your reference image, use Nano Banana Pro instead of GPT-image 2, nano was trained on older phones so it's not overpolished 3. get a clear image of the product you want to advertise 4. upload everything you have to GPT with prompt template (below) to generate prompt 5. upload all 3 references to Seedance 2.5 and generate using references mode, 720p is enough, and your preferred aspect ratio and duration 6. watch the result with sound and always check the eyes, skin, voice, and background 7. try the same setup in Wan 3.0 for a 2x cheaper alternative === prompt template: You are an expert at writing prompts for Seedance 2.5 (reference mode) that produce realistic, TikTok-style UGC videos. Nothing about the result should look or sound like AI. I'm uploading: - @Image1 – the exact first frame of the video (the person, outfit, setting, lighting and framing) - @Video1 – a real TikTok video. Copy the person's physical performance, energy, movement and way of talking from it - @Image2 – the product Study all three files carefully. Then write one prompt in plain English, lowercase, with these 7 numbered sections and nothing else: 1. format & visual references The length of the video, "natural tiktok ugc style", @Image1 as the exact first frame and visual reference for the person, outfit, setting, lighting and framing, @Video1 as the reference for physical performance, voice and accent. 2. body language & movement Match the posture, hand gestures, facial movements and body language from @Video1. Describe 1–2 specific movements you can actually see in @Video1 (for example, how she sits down, leans in or moves her hands while she talks). 3. dialogue "she says exactly what is written between the brackets: [ ... ]". Write the script in the brackets. It should sound like a real person talking to a friend, not like an ad: short sentences, small reactions, no marketing words. It must fit the video length when spoken at a natural pace. 4. voice, emotions & reactions How she speaks (pace, energy, tone) and how her emotions change line by line. Quote specific lines from the script and describe the reaction on each one (a quick laugh, disbelief, getting excited, turning mock-serious, ending warmly). Ask for natural changes in pitch, emphasis and speed, with tiny pauses that fit the reactions. 5. product interaction Exactly when and how @Image2 comes into the video (where she takes it from, on which line it appears), how she shows it to the camera with the label visible, and what her other hand does at the same time. 6. timing & realism Fit the full performance within the video length. Realistic skin texture, natural blinking, accurate lip sync and room sound that fits the setting (echo in a bathroom, quiet hum in a bedroom, wind and traffic outside). 7. camera & restrictions One continuous shot, the same camera as in @Image1 (handheld or stationary), and a list of what the video must NOT have: no shouting, no music, no cuts, no zooms, no captions, no tiktok logo, no extra spoken words. Rules: - Use @Image1, @Video1 and @Image2 exactly like that, so Seedance knows which file is which. - Describe only what you can see in the files. Don't invent a new room, outfit or person. - Keep every section short and concrete. No general phrases like "high quality" or "cinematic". - Give me only the finished prompt, ready to paste. MY VIDEO - Product: [product name + what it does] - Video length: [e.g. 15–20 seconds] - Accent: [e.g. american] - What she should say / the idea: [e.g. she just spent the last of her money on this cream and she's excited but feels a bit guilty]

Friday was my last day at Anthropic. Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env. I steered it for tens of hours but never manually wrote a line of code. It worked, I was hooked, and I joined the team. When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests.  Today, the default is trusting Claude's judgment to auto-approve. Claude does large migrations and builds new products.  Helping build Claude Code into an industry-defining product has been the journey of a lifetime.  Thank you to the incredible team behind Claude Code, Cowork, and Tag. These products are the sum of thousands of great ideas from you. I'll miss working with you all. Thank you to our users for embracing a new way of working. Your constant feedback shaped so much of what we shipped and your stories made the work so meaningful. Excited to see what you all make next!