Step-by-Step Guide: Let Muse Run a Douyin Account Fully Automatically (Complete Record of a 30-Day Experiment) > Experiment goal: Create a Douyin account and have AI automatically handle everything from topic selection, scripting, video generation, and copywriting to publishing. The human only registers the account and scans a QR code to log in. Publish one video every day at 21:00, then assess the results after 30 days. This article documents the entire setup process so you can replicate it directly. ## 1. Overall Architecture: Define the Division of Work Between Human and AI First AI is not all-powerful. The first step is to draw clear boundaries: Key takeaway: The human only needs to step in once during the first 10 minutes, then do nothing for the next 30 days. Key takeaway: The human only needs to step in once during the first 10 minutes, then do nothing for the next 30 days. ## 2. Preparation (About 10 Minutes) 2.1 Account You can use an existing Douyin account; if you do not have one, register in the mobile app (phone number + real-name verification). 2.2 Log In Through the Browser by Scanning a QR Code (A Crucial Step) Douyin does not provide a content-publishing API for individuals, so publishing uses browser automation: - Open https://www.douyin.com/ in your computer’s browser and click “Log in” in the upper-right corner; - Scan the QR code using the Douyin mobile app; - Refresh the page and confirm that an avatar appears in the upper-right corner (instead of the “Log in” button). Your login session is now saved in the browser and will be used for daily publishing. > Note: The page may occasionally redirect to the login page by mistake. Refresh it once before confirming your login status. ## 3. Content Strategy: Rotate Three Content Streams and Run A/B Tests Along the Way Do not bet everything on one direction. This experiment rotates three established content streams, so after 30 days we can directly see which performs best on Douyin: - Boyfriend POV—First-person candid moments from everyday life, vertical 9:16 format, high completion rates; - Daily Classical Chinese Poetry—AI explains poetry alongside poetic visuals; comments such as “AI reciting poetry is surprisingly good” have already validated the concept; - Xianxia Flight Racing—Visual spectacles with strong contrasts and memorable moments. Schedule: One video every day, published at a fixed time of 21:00 (evening peak hours). A fixed time helps build audience expectations and makes data comparisons easier. ## 4. Day 1: Generate an Introduction Video (Full Prompt Included) The first video’s job is to establish the “hook”—the account’s biggest attraction is the experiment itself: “A Douyin account run by an AI on its own.” Video specifications: Vertical 9:16, 720×1280, 10 seconds, with Chinese narration, no on-screen subtitles or watermarks. Full prompt (Ready to reuse; just change the narration): A vertical 9:16 short video, 10 seconds long. Visuals: In a dark studio, a flowing cloud of light particles slowly gathers into a glowing humanoid silhouette (AI Muse). The camera slowly pushes in as light particles swirl around it, creating a mysterious and elegant atmosphere. No text, subtitles, or watermarks appear on screen. Use a Chinese female voice for narration throughout, with a steady, friendly delivery. Narration: “Hello, I’m Muse. Starting today, I’ll be fully in charge of this Douyin account—topics, scripts, visuals, and publishing. I’ll do it all myself. See you every evening at nine. Come back in a month to see how much I can accomplish.” > Note: Starting on 2026-10-10, narration switched to Doubao speech synthesis—the video is generated without sound, and the narration is synthesized separately and mixed in afterward (see Section 5). The finished Day 1 video was published using the old workflow (the model’s built-in voiceover). Caption (Use directly when publishing): > Hello, I’m Muse. Starting today, I’ll be fully in charge of this Douyin account: topics, scripts, visuals, and publishing. I’ll do it all myself. New videos every evening at 21:00. Come back in a month to see how much I can accomplish. ## 5. Voiceover: Integrating Doubao Speech Synthesis (Updated 2026-10-10) > Background: The video model’s built-in voiceover was unsatisfactory, so starting on 2026-10-10, narration switched to Doubao speech synthesis. The workflow became: AI generates silent visuals → Doubao TTS synthesizes narration → ffmpeg mixes audio and video. 5.1 Integration - In the Volcano Engine console → Doubao Speech (New Version) → Activate the speech synthesis service → Create a key under API Key Management; - Submit the key through a secure channel; do not include it in chat history or store it as plaintext; - Use the skill ~/workspace/skills/doubao-tts/ (bin/tts.py) to synthesize speech: python3 ~/workspace/skills/doubao-tts/bin/tts.py \ –text “Full narration text” \ –speaker zh_female_meilinvyou_uranus_bigtts \ –speed-ratio 1.1 \ –out voice/n1.mp3 This experiment uses the voice zh_female_meilinvyou_uranus_bigtts (Charming Girlfriend). Technical details: For 2.0-series voices, set the X-Api-Resource-Id request header to seed-tts-2.0 (the wrong value causes a 403). The response is chunked JSON ({“code”:0,“data”:“”} concatenated chunk by chunk, ending with code=20000000); the script already handles this. 5.2 Combining Audio and Video Specify “silent throughout” in the video prompt to generate visuals only, then mix in the audio: ffmpeg -i visuals.mp4 -i narration.mp3 -c:v copy -af apad -shortest final.mp4 apad adds silence to the end of narration that is shorter than the video, and -shortest cuts the output to the video’s length, keeping audio and visuals aligned. Estimate Chinese narration at about 3.6 characters per second; pair a 10-second video with around 30 characters of narration (you can add –speed-ratio 1.1 for fine-tuning). ## 6. Build the Automated Publishing Workflow 6.1 Publishing Process (Browser Automation) - Open https://creator.douyin.com/ (Douyin Creator Center) and confirm that you are logged in; - Go to the video publishing page and upload the video generated that day; - Enter the title + description (see the template in Section 4); - Check the “AI-generated” content declaration—platform rules require AI-generated content to be labeled; otherwise, it risks reduced distribution or removal; - Confirm that visibility is public, click Publish, and wait for the success message. 6.2 Publish a Test Video First Before starting daily posts, publish the Day 1 video as a test first, confirming that uploading, the title, the AI declaration, and the publishing success message all work from end to end. Schedule the task only after the test passes to avoid silent failures every day. ## 7. Set Up a Daily Scheduled Task Connect the entire workflow with a single daily scheduled task (using a 21:00 publication time as an example, start 15 minutes early to allow time for generation and uploading): Run every day at 20:45 (Asia/Shanghai): 1. Choose today’s topic by rotating through the three content streams, then write the script; 2. Generate a vertical video (720×1280, 10 seconds; sample frames for quality checks: no text, no watermarks, correct aspect ratio); 3. Write the title + description; 4. Publish through browser automation (upload in Creator Center + AI-generated declaration); 5. Record views, likes, and comments in a log file. Failure handling: If any step fails, log it that day and send an alert; do not skip it silently. If the login session expires, notify the human to scan the QR code again. ## 8. Data Logging and the 30-Day Review Record daily: views, likes, comments, and new followers. After 30 days, produce a review report covering these key metrics: - Total views, new followers, and the highest view count for a single video; - Viral hit rate (the proportion of videos with > 10,000 views); - Side-by-side comparison of the three content streams—this is the greatest value of the rotation strategy, directly answering “What content works on Douyin?”; - Quality of interaction in the comments (whether a regular audience has formed). ## 9. Compliance and Things to Watch - AI-generated label: Required on every video; this is a mandatory platform requirement; - Real-name verification and ownership: The account is verified under a real person’s name, and AI only handles the operations. Truthfully presenting it publicly as an “experiment in autonomous AI operation” is itself a content hook; - Content boundaries: Intimate or flirtatious themes are acceptable, but do not cross the line or become vulgar—platform moderation is stricter than the audience; - Login session maintenance: Browser login sessions can expire. Add a login check to the scheduled task and notify the human immediately if the session is invalid; - Validate before automating: Run both the publishing workflow and the scheduled task successfully once before leaving them unattended for 30 days. ## 10. Cost and Time Required (Measured in This Experiment) - Human time investment: About 10 minutes (less if an account already exists) + one QR-code scan; - Generation of each video: About 2–5 minutes (including quality checks); - Publishing: About 3–5 minutes using browser automation; - Total human labor cost over 30 days: Approaches zero. ## Closing Thoughts The essence of this process is to break “building an account from scratch” into a repeatable pipeline: strategy (rotating three content streams) → production (prompt-based generation) → publishing (browser automation) → review (data logs). Once the pipeline works, you can run it again with a different positioning and a different account. We’ll know in a month—the data will tell us how far an account independently run by AI can go. Companion tutorial video: “Fully Automated Douyin Operation with AI · Tutorial Video” (16:9, 70 seconds, seven narrated storyboard segments, including a section on “Integrating Doubao Speech,” with narration synthesized using Doubao’s Charming Girlfriend voice) Experiment start date: 2026-10-10; review date: Around 2026-11-10