A streamlined process from idea, script, character design to a consistent ASMR image and video creation prompt — all with ChatGPT and Google Flow.
Animated ASMR videos are captivating not only because of their beautiful visuals. What keeps viewers engaged is a world with its own rhythm: the sound of rain hitting the roof tiles, dough being kneaded slowly, firelight reflecting on pottery, and a small story told without dialogue.
In this tutorial, we'll build the entire process using just two tools:
• ChatGPT: Develop ideas, "series bible," scripts, shot lists, and prompts. • Google Flow: Create reference images, generate motion clips, and arrange scenes into videos.
The goal isn't to copy a famous movie or character. We'll create an original world, using a warm, rich, hand-drawn animation visual language that has the potential to develop into a long-running series.
The finished product we will create
A recurring example throughout the article is the "Rainy Moon Bakery" series:
• Main character: An, a 24-year-old baker with black hair tied in a low bun and a moss-green apron. • Companion: Mun, a small, charcoal-colored cat with amber eyes. • Setting: A wooden bakery on the edge of a forest, operating on rainy nights. • Emotion: Warm, peaceful, healing, no dialogue. • Sounds: Rain, fire, pottery, pastry wrapping paper, footsteps, and the very soft meowing of the cat.
First video: “A rainy night at the bakery: An makes milk tarts for a late customer.”
Why are AI videos often disjointed?
The three most common mistakes are:
1. The characters change faces, clothes, or proportions between scenes. 2. Each prompt uses a different style, making the video look like a collection of random clips. 3. The scenes have too much action, resulting in choppy movement and chaotic sound.
The solution is to lock the three layers before creating any images:
• Identity Anchor: A fixed character description. • Visual Anchor: Fixed color palette, textures, lighting, and camera language. • Audio Anchor: A consistent set of sounds throughout the film and those that are not to be shown.
7-step process
Step 1: Choose a small “ASMR ritual”
Don't start with a grand storyline. Choose an activity with easily recognizable sounds and movements:
• Baking cakes on a rainy night. • Brewing tea by the window in winter. • Repairing pottery in a sunny room. • Preparing lunch boxes for an early train journey. • Caring for a tiny garden after a storm.
Each video should have only one emotional goal and one central action.
Step 2: Use ChatGPT to create a “bible series”
The Bible series is the original source material that ensures all videos share the same world, characters, and feel. Once created, save it and include it at the beginning of each new work session.
Master Prompt 1 — Building an original concept series
Prompt mẫu
You are a creative director specializing in developing wordless ASMR animation series. Build an original series concept based on the following information: - Audience: [TARGET AUDIENCE] - Main Emotion: [HEALING / WARMTH / FANTASTIC / NOSTALGIA] - Central ASMR Activity: [BAKING / TEA MAKING / POTTERY MAKING / GARDENING] - Video Duration: [MINUTES] - Aspect Ratio: [16:9 or 9:16] Required Requirements: 1. Create a completely original world, characters, and story. 2. Do not use copyrighted studio names, films, or characters as stylistic indicators. 3. No dialogue, narration, or on-screen text. 4. Each episode must have a clear everyday ritual and a small emotional shift. 5. The design must be simple enough to keep the character consistent across multiple AI scenes. Return structure: A. Series title and content promise in one sentence B. World description C. Profiles of up to 2 main characters D. Fixed visual anchor E. Fixed audio anchor F. 10 first-episode ideas, each including visual hook, ASMR action, and emotional ending G. List of elements that absolutely do not change between episodes
Step 3: Lock character and style
After ChatGPT returns the results, request that it be condensed into a descriptive block that can be repeated verbatim in every prompt.
Master Prompt 2 — Create Identity Anchor and Visual Anchor
Prompt mẫu
Based on the Bible series below, create a consistent key set for use with the AI image/video creation tool. [PASTE BIBLE SERIES] Return the following 5 parts: 1. CHARACTER_ANCHOR: - Write a 70–110 word English paragraph. - Describe the adult age, face, hair, clothing, colors, body proportions, and an identifying prop. - Only describe visible details. 2. COMPANION_ANCHOR: - Write a 30–60 word English paragraph for a supporting character or companion animal. 3. WORLD_ANCHOR: - Describe the architecture, materials, weather, and objects always present in the setting. 4. VISUAL_ANCHOR: - Original hand-drawn animation style, soft watercolor, tangible pencil strokes, soft cinematic lighting, handcrafted materials. - Do not mention the artist, studio, film, or brand name. 5. NEGATIVE_ANCHOR: - no text, no logo, no watermark, no extra fingers, no duplicate character, no costume change, no face drift, no modern plastic objects, no abrupt lighting change The anchors must be ready to be copied verbatim into every subsequent prompt.
For this sample project, Visual Anchor could be:
Prompt mẫu
Original cozy hand-painted animated film aesthetic, soft watercolor backgrounds, visible graphite linework, tactile wood and ceramic textures, warm amber practical light against deep indigo rain, restrained natural motion, cinematic composition, peaceful handcrafted atmosphere, consistent character proportions, no text, no logo, no copyrighted characters.
Step 4: Write the script according to the scenes.
An AI clip should focus on one main action. Instead of asking "An opens the door, runs into the kitchen, kneads the dough, and bakes the cake," break it down into four short scenes.
Master Prompt 3 — ASMR Scripts and Shot List
Prompt mẫu
You are an animation director and sound designer for a wordless ASMR video. Project Information: - Episode Title: [EPISODE TITLE] - Target Duration: [DURATION] - Number of Scenes: [NUMBER OF SCENES] - Aspect Ratio: [16:9 or 9:16] SERIES BIBLE: [PASTE SERIES BIBLE] IDENTITY + VISUAL + AUDIO ANCHOR: [PASTE ANCHORS] Write a structured story: 1. Visual hook in the first 5 seconds 2. Preparation 3. Main ASMR ritual 4. A small, gentle hiccup 5. Resolution moment 6. A concluding scene that feels complete and can be smoothly repeated Rules: - No dialogue, no narration, no text on screen. - Each scene has only one main action and a maximum of one camera movement. - Prioritize close-ups of hands, materials, steam, rain, fire, and pottery. - Movement should be slow, weighty, and physically realistic. - Do not add characters outside the Bible series. Return the following table: SCENE_ID | DURATION | STORY_BEAT | SUBJECT_ACTION | ENVIRONMENT | SHOT_SIZE | CAMERA | LIGHTING | ASMR_AUDIO | TRANSITION | CONTINUITY_NOTES After the table, add: - A summary sentence of the video's emotional rhythm - The three most important scenes to create reference shots for beforehand - A list of sounds that must remain consistent throughout
Step 5: Convert the shot list into an image creation prompt in Flow.
Create keystone images for important scenes first. Once you have the desired character and background images, use them as references for subsequent scenes if the Flow mode you're using supports keystone images.
Master Prompt 4 — Prompt for creating images for Google Flow
Prompt mẫu
You are the prompt designer for Google Flow. Task: Convert the shot list into a consistent image creation prompt. The final prompt must be written in English, while the explanatory notes should be written in Vietnamese. CHARACTER_ANCHOR: [PASTE EXACTLY AS ONE] COMPANION_ANCHOR: [PASTE EXACTLY AS ONE] WORLD_ANCHOR: [PASTE EXACTLY AS ONE] VISUAL_ANCHOR: [PASTE EXACTLY AS ONE] NEGATIVE_ANCHOR: [PASTE EXACTLY AS ONE] SHOT LIST: [PASTE SHOT LIST] For each scene, return: 1. SCENE_ID 2. KEYFRAME_PROMPT — a 120–180 word English text in the following order: subject identity → single frozen action → environment → foreground/midground/background → shot size → camera angle → lighting → color → material texture → continuity → aspect ratio 3. NEGATIVE_PROMPT 4. REFERENCE_NEEDED — specify Use character images, background images, or both. 5. CONSISTENCY_CHECK — 3 details that must be compared with the previous scene. Do not use vague phrases like “beautiful,” “masterpiece,” or “best quality” unless they describe it with visible details. Do not mention the studio, film, artist, or copyrighted characters.
Step 6: Create motion and sound prompts in Flow.
A good video prompt should clearly state: what is moving, how it is moving, what the camera is doing, and which sounds are coming from near the microphone.
Master Prompt 5 — Image-to-video/text-to-video prompt for Flow
Prompt mẫu
You are an AI film director specializing in writing short video prompts with natural motion and ASMR sound. INPUT: - Shot list: [PASTE SHOT LIST] - Fixed anchors: [PASTE ANCHOR] - Approved image/keyframe prompt: [PASTE PROMPT OR IMAGE DESCRIPTION] Create a VIDEO_PROMPT in English for each scene. Required structure for each prompt: A. Continuity: Characters, costumes, props, and lighting must remain consistent from the keyframe. B. Primary motion: Only one main action, depicting speed, direction, and weight. C. Secondary motion: A maximum of two very subtle environmental movements. D. Camera: A simple motion or static camera. E. Physics: Depicting material reactions such as powder, steam, fabric, rain, or fire. F. Audio: Close-up sound, distant background sound, and silences. G. Ending frame: The ending pose helps transition to the next scene. H. Avoid: Motion errors and unwanted elements. Rules: - No dialogue, no narration, no background music. - No jump cuts within the same clip. - No changes to faces, costumes, props, or architecture. - No adding new people or animals. - No fast camera movements, strong shaking, or sudden zooms. - Sound must be realistic and correspond accurately to the material being shown. Return table: SCENE_ID | VIDEO_PROMPT | FIRST_FRAME | END_FRAME | AUDIO_PRIORITY | RETRY_HINT
Step 7: Pre-publication review
Don't recreate the entire video when only one scene is faulty. Use ChatGPT to pinpoint the exact prompt causing character drift, physics errors, or excessive action, and then only recreate that specific scene.
Master Prompt 6 — Check quality and fix error scenes
Prompt mẫu
You are the continuity supervisor and quality editor for AI videos. BIBLE SERIES + ANCHORS: [PASTE DOCUMENT] ORIGINAL SHOT LIST: [PASTE SHOT LIST] DESCRIPTION OF EACH CLIP'S RESULTS: [PASTE DESCRIPTION OR ERROR NOTES] Check according to 8 criteria: 1. Character faces and proportions 2. Costumes and props 3. Movement direction 4. Lighting and time of day 5. Spatial positioning 6. Logic of physics 7. ASMR rhythm and sound clarity 8. Ability to connect first frame/end frame Return table: SCENE_ID | PASS/RETRY | SPECIFIC ERROR | CAUSE IN PROMPT | CORRECT PROMPT QUESTION | DO WE NEED TO CHANGE THE KEYFRAME? We only recommend recreating scenes that are genuinely faulty. We will not change the locked character and world designs.
Sample prompts: “Rainy Moon Bakery”
Scene 01 — Hook outside the bakery
Image prompt:
An original hand-painted animated film frame of a tiny timber bakery glowing at the edge of a rain-soaked indigo forest at night. Warm amber light spills through two small windows onto wet stones. An adult Vietnamese baker named An is visible only as a calm silhouette behind the window, her black hair tied in a low bun and her moss-green apron clearly readable. A small charcoal cat with amber eyes sits beneath the awning. Soft watercolor foliage, visible graphite linework, tactile wood grain, gentle rain haze, cinematic wide establishing shot from slightly below eye level, peaceful handcrafted atmosphere, deep indigo and warm amber palette, no text, no logo, no copyrighted characters, 16:9.Prompt video:
Keep the bakery, window layout, character silhouettes and color palette identical to the reference image. Fine rain falls steadily and creates small circular ripples in the foreground puddles. The charcoal cat slowly turns one ear toward the bakery door; this is the only character action. Warm window light flickers almost imperceptibly. Static cinematic camera with a very slow two-percent push-in. Close rain taps on the wooden awning, distant soft thunder, quiet fire crackle from inside, no speech, no music. End with the cat looking toward the door. Preserve faces, architecture and lighting; avoid fast rain, camera shake, new objects or additional characters.Scene 02 — Kneading the dough
Prompt video:
Close-up of An's adult hands slowly folding soft milk-bread dough on a worn wooden worktable. Keep her moss-green apron, cream rolled sleeves, brass measuring spoon and warm amber side light unchanged. One continuous kneading cycle: press, fold, quarter-turn, rest. Flour particles move only when touched; the dough compresses with realistic weight and elasticity. Locked overhead camera. Intimate sounds of palms against dough, faint wooden table creak, rain muffled behind the window and low oven fire, no dialogue, no music. End with both hands resting beside the rounded dough, ready for the next shot. Avoid extra fingers, melting dough, fast motion, floating flour or changes to the utensils.Scene 03 — Mun touches the bag of flour
Prompt video:
The small charcoal cat Mun, with consistent amber eyes and a tiny nick on the left ear, reaches one paw toward a folded paper flour bag beside the worktable. The paw gently touches the bag once; the paper crinkles and a tiny dusting of flour falls onto the wooden floor. An remains softly out of focus in the background and does not change pose. Low floor-level camera, static composition, shallow depth of field, warm bakery light and cool rainy window light. Close paper crinkle, one soft paw step, distant kneading and rain, no speech, no music. End with Mun looking at the small flour mark. Avoid duplicate paws, jumping, bag deformation, face drift or new objects.Scene 04 — The cake rises in the oven
Prompt video:
Macro view through the small oven window as six milk buns slowly rise and turn golden. Keep the same ceramic baking tray and oven interior from the reference frame. The primary motion is a subtle, physically plausible expansion of the dough; secondary motion is gentle heat shimmer and a quiet pulse of orange firelight. Camera remains locked. Detailed oven crackle, a soft metal tick as the tray warms, muffled rain in the distance, no speech, no music. End on the fullest golden shape of the buns. Avoid rapid inflation, burning, liquid texture, moving tray or abrupt lighting changes.Scene 05 — Repeatable ending
Prompt video:
An places one warm milk bun on a small blue-gray ceramic plate beside the rainy window. Keep her adult face, low black hair bun, moss-green apron and all counter objects identical to previous shots. Steam curls upward in one slow ribbon while Mun settles beside the plate without touching it. Very slow side dolly of a few centimeters, ending on the window where rain reflections match the opening scene. Close ceramic contact, delicate paper wrap rustle, soft purring, rain on glass and low fire crackle, no dialogue, no music. End with the same amber window glow and rain rhythm used in Scene 01 to support a seamless loop. Avoid eating, sudden gestures, extra food, text, logos or character drift.Master Prompt 7 — Titles and thumbnails aren't overly sensationalized.
Prompt mẫu
You are the YouTube packaging strategist for a series of animated, non-dialogue ASMR videos. Episode Content: [PASTE EPISODE SUMMARY] Create: 1. 15 Vietnamese titles 2. 15 natural-sounding English titles (not machine translations) 3. 5 well-structured concept thumbnails in small sizes 4. 3 video descriptions, each under 160 characters Criteria: - Create intrigue through situations and feelings, without promising income. - Avoid using “100%”, “secret”, “superb”, “automatic monetization”, or unverified data. - Titles should be under 65 characters if possible. - Thumbnails should feature only one main subject, one action, and one contrasting highlight. - Do not copy posters, characters, or the visual identity of famous films. Rate each title on a scale of 10 based on: clarity of content, emotional impact, originality, and readability at a quick glance. Select the three best options and provide a brief explanation.
Checklist before posting
Checklist thực hành
- The characters maintain accurate facial features, hair, clothing, and proportions.
- Each clip features only one main action.
- No text, logos, or AI-generated watermarks.
- The gaze and prop placement match between two consecutive scenes.
- The sound is authentic and free of unexpected voices.
- The opening scene has a visual hook in the first few seconds.
- The closing scene creates a sense of completion or a smooth transition.
- The title and thumbnail accurately reflect the content.
- The video has its own story, creative choices, and editing.
- Do not use characters, logos, music, or frames from other works.
Reusable formula
You don't need to create an "automated video machine." A better system is:
Một thế giới nguyên bản
→ một series bible ổn định
→ một nghi thức ASMR cho mỗi tập
→ một hành động cho mỗi clip
→ khóa first frame/end frame
→ kiểm tra thủ công
→ chỉ tạo lại cảnh lỗiOnce this foundation is solid, ChatGPT helps you think structured, while Flow transforms that structure into visuals and movement. The real value still lies in creative choices, storytelling rhythm, and the ability to keep each episode with its own unique emotion.
