Workflow + Prompt · มีพื้นฐาน
Google Flow Workflow — Coffee Knight → Dragon Reveal
อย่าเริ่ม AI Video ด้วยการสุ่มวิดีโอ — ล็อก Character, Location และ Assets ก่อน แล้วค่อย animate ทีละ shot
แก้ปัญหาผลิตวิดีโอ AI ให้นิ่ง
- Prompt คัดลอกได้ 8
- เช็กลิสต์ 12 ข้อ
ใช้อะไรบ้าง
- Image generation: ChatGPT image generation — อันนี้คือวิธีที่เราใช้ทำ character sheet / location / dragon / anchor frame
- Alternative: ถ้าใครถนัด Google จะใช้ Nano Banana แทน ChatGPT สำหรับภาพก็ได้
- Video generation: Google Flow — ใช้ทำแต่ละ shot แยกกัน ไม่ยัดทุกอย่างลง prompt เดียว
- Editing: CapCut / Premiere / Final Cut / editor ที่ถนัด สำหรับ trim, cut, rotate 90°, sound design
STEP 1 — สร้าง ASSETS ให้ทุกอย่างนิ่งก่อน
อย่าเริ่มจากวิดีโอทันที เพราะถ้ายังไม่มี character / dragon / location lock โมเดลจะเปลี่ยนหน้า ชุด ขนาดมังกร และฉากไปเรื่อย ๆ
ลำดับ asset ที่เราทำ:
- Mew Character Sheet
- Foggy Field Location
- Dragon Design
- Scene Anchor — Mew ยืนหน้ามังกร
1.1 Character Sheet — Mew Knight
Tool: ChatGPT image generation
Alternative: Nano Banana
ใส่อะไร: รูปหน้าตัวเอง + รูปชุด/armor reference ที่อยากได้
เป้าหมาย: ได้ sheet เดียวที่ล็อก หน้า / ผม / ชุด / ดาบ / cape ให้ใช้ซ้ำทุก scene
ตัวอย่างภาพที่เราได้

เอาภาพ character sheet ของ Mew มาแปะตรงนี้
Prompt — Character Sheet
SCENE CONTEXT
Character reference turnaround sheet of one young male knight in three full-body views (front, profile, back) plus one close frontal portrait, for identity and costume lock.
ACTIVE REFERENCES
<<<image_1>>> — primary identity reference: controls Mew’s face, skin tone, eye color, facial proportions, hairstyle feeling, and overall likeness. 100% match.
<<<image_2>>> — supporting identity / profile reference: controls profile-only anatomy when available — nose bridge and tip shape, jaw and chin curve, lips seen in profile, ear shape.
<<<image_3>>> — primary costume reference: controls the knight armor, fleur-de-lis chest motif, engraved silver plate, chainmail placement, leather belt rigging, sword placement, and cloak silhouette.
CHARACTER ANCHORS
Mew is a 23-year-old Thai / East Asian man, 170 cm tall, slim-light build, not bulky, not overly muscular.
Face: youthful soft oval face, natural jaw, dark brown almond-shaped eyes, dark natural brows, straight narrow nose bridge, natural lips, no facial hair.
Skin: natural warm light-beige skin with real visible skin texture, no plastic smoothing, no westernization.
Hair: dark brown-black, medium-length tousled layered hair, slightly messy natural volume, never blonde or auburn.
ARMOR AND COSTUME
High-polish cool white-silver plate armor with engraved fleur-de-lis detailing.
Blackened chainmail visible at joints and underlayers.
Brown leather belts with aged buckles.
A sheathed sword hangs clearly at the RIGHT HIP.
Full-length ivory-white cloak with real fabric weight and structured drape.
BODY AND PRESENTATION LOCK
170 cm, slim-light build, natural male proportions.
No bodybuilder anatomy.
No oversized fantasy armor.
No helmet.
SHEET LAYOUT
Single reference sheet on neutral seamless light-grey background.
Panel 1 — full body front view, neutral stance.
Panel 2 — full body profile view, same proportions, sword still sheathed at RIGHT HIP.
Panel 3 — full body back view, full cloak visible.
Panel 4 — close frontal portrait, shoulders-up, calm neutral expression.
LIGHTING
Soft even neutral studio lighting, low contrast, accurate material/color reproduction.
POSITIVE CONSTRAINTS
Same face, same armor, same cloak, same hair, same body proportions across all panels.
Armor remains cool white-silver.
Sword remains on RIGHT HIP.
No extra character.
No extra props.
No text, logo, watermark.
Photorealistic production-ready reference sheet.1.2 Location Plate — Foggy Field
Tool: ChatGPT image generation
Output: 16:9
ห้ามมี: คน / มังกร / อาคาร
เหตุผล: เราต้องมีฉากกลางที่ reuse ได้ ไม่ให้ Flow เปลี่ยนภูมิประเทศทุกคลิป
ตัวอย่างภาพที่เราได้

เอาภาพ foggy field 16:9 มาแปะตรงนี้
Prompt — Location
Photorealistic location plate, horizontal 16:9, full realism — a wide, air-filled frame: a vast empty field of tall grass dissolving into dense fog, dominated by luminous grey emptiness. No characters, no animals, no structures.
COMPOSITION
Grass field occupies only the LOWER THIRD of frame.
Upper TWO THIRDS is open grey-white fog-space with no hard horizon.
Wide quiet composition with lots of negative space for later staging.
FIELD
Dense knee-to-thigh-height grass, natural green mixed with dry yellow stalks and seed heads, wet with fog-dew.
STONES
A few sparse dark weathered boulders half-sunk in grass, wet with moss and pale lichen.
FOG
Dense even fog wall, visibility around 40–60 meters.
No hard horizon line.
Background dissolves naturally into grey-white void.
CAMERA
Locked wide documentary-like frame, slightly above head height, level horizon.
LIGHT
Flat soft shadowless overcast-fog light, cool-neutral grey-white, muted natural greens.
REALISM
True grass detail, wet rock, volumetric fog.
No illustration, no painterly look, no CG sheen, no oversaturation.
CONSTRAINTS
No people.
No dragon.
No buildings.
No paths.
No fences.
No text, logo, watermark.1.3 Dragon Asset — Colossal White Dragon
Tool: ChatGPT image generation
Alternative: Nano Banana
เป้าหมาย: ล็อก species, head shape, wing shape, scale texture, size feeling ก่อนเอาเข้า Flow
ตัวอย่างภาพที่เราได้

เอาภาพ dragon reference / character sheet มาแปะตรงนี้
Prompt — Dragon Asset
SCENE CONTEXT
Creature reference sheet of one colossal white dragon for identity and design lock.
DESIGN GOAL
One single consistent colossal wild white dragon.
Ancient, regal, terrifying, enormous.
Physically real, heavy, mass-driven.
Not cartoonish, not gamey, not a pet, not a mount.
DRAGON ANCHORS
Massive four-legged winged dragon with folded wings, long neck, powerful shoulders, huge torso, enormous head.
SCALE
Absolutely colossal compared with a 170 cm human.
Its head alone should feel many times larger than a person.
HEAD
Long heavy reptilian-draconic skull, horned ridge structure, layered facial scales, pronounced brow, strong jaw.
SURFACE
White to white-grey overlapping keratin scales with growth ridges, chipped edges, dirt in seams, matte-to-soft-satin natural finish.
No plastic gloss.
WINGS
Massive folded wings with pale leathery membranes and long finger bones.
Wing membrane is a different softer material from body scales.
BODY FEEL
Ancient, heavy, slow, enormous inertia.
NO TACK
No saddle.
No reins.
No bridle.
No straps.
SHEET LAYOUT
Multiple clean views on neutral light-grey background:
1. full body 3/4 view
2. side view
3. close-up head portrait
4. rear / folded-wing view
Same dragon identity in every panel.
No human.
No environment storytelling.
Photorealistic production reference sheet.1.4 Scene Anchor — ก่อนทำวิดีโอจริง
นี่คือขั้นที่ช่วยลดการมั่วมากที่สุด
ก่อนเจน scene มังกรแบบ video ให้สร้าง ภาพนิ่ง composition ที่ถูกก่อน แล้วค่อยใช้เป็น visual target/reference
Tool: ChatGPT image generation
Input: Mew character sheet + dragon asset + foggy field location
Output: 16:9
ตัวอย่าง frame ที่เราใช้เป็น anchor

เอาภาพ Mew ยืนหน้ามังกรมาแปะตรงนี้
Prompt — Scene Anchor
SCENE CONTEXT
Photorealistic live-action fantasy anchor frame for the knight-versus-dragon confrontation.
ACTIVE REFERENCES
<<<image_1>>> — Mew character sheet: controls identity, armor, cloak, hair and RIGHT-HIP sword placement.
<<<image_2>>> — dragon reference: controls dragon anatomy, texture, head, wings and scale.
<<<image_3>>> — foggy field location: controls environment and atmosphere.
COMPOSITION
Horizontal 16:9.
Low camera position in the field.
Mew stands relatively small in frame, seen from behind / back-three-quarter.
He faces the dragon.
The dragon dominates most of the background and feels overwhelmingly huge.
STORY STATE
Dragon is asleep.
Head resting low on the ground.
Eyes closed.
Wings folded.
Slow heavy breathing.
Mew stands still.
He is not walking toward it.
His sword remains sheathed on the RIGHT HIP.
SCALE
Dragon is absolutely colossal.
Its head alone is enormous compared with Mew.
Mew must read as a tiny human against its bulk.
ATMOSPHERE
Foggy overcast daylight, wet grass, muted natural palette, epic silence.
NO action yet.
NO roar.
NO fire.
NO sword draw.
NO walking.STEP 2 — ทำ VIDEO ใน GOOGLE FLOW
ตอนนี้เรามี asset ครบแล้ว ถึงค่อยทำ video เป็น หลาย shot สั้น ๆ แล้วเอามาต่อกัน
Shot Map ที่เราใช้
- Scene 1 — Coffee selfie
- Transition A — Camera launch
- Transition B+C — High tumble → fall → ground impact
- Orientation cut — portrait → landscape
- Scene 2 — Dragon reveal + sword draw + eye open
2.1 Scene 1 — Coffee Selfie
Google Flow settings
- Ratio: 9:16
- Duration: 10s ถ้าโหมดที่ใช้รองรับ
- References: Location + Mew character sheet + Mew face/reference
- ไม่ใส่ dragon
ตัวอย่างเฟรมท้ายที่เราใช้ต่อ transition

Prompt — Scene 1
SCENE 1 — COFFEE BEFORE THE DRAGON
One continuous realistic phone-POV shot, vertical 9:16, exactly 10 seconds.
MEW, a young East Asian male knight, 170 cm tall, stands alone in a foggy grass field wearing white-silver engraved armor and an ivory-white cloak.
The recording viewpoint is held naturally in his LEFT hand.
THE RECORDING DEVICE ITSELF IS NEVER VISIBLE.
His RIGHT hand holds a clear plastic cup of iced coffee with transparent lid, straw, visible ice cubes, dark coffee, and natural condensation.
Mew does NOT look into the lens.
He looks FORWARD and SLIGHTLY UPWARD into the fog, toward something enormous off-screen ahead.
PERFORMANCE
Calm, deadpan, confident, slightly tired, unimpressed.
No smile, no panic, no influencer eye contact.
0:00–0:03 — THE STARE
Mew stands still, looking forward and slightly upward into the fog.
Wind moves his hair and cloak naturally.
0:03–0:06 — THE SIP
Without changing his gaze, he raises the iced coffee with his RIGHT hand and takes a long relaxed sip through the straw.
Ice shifts and clinks naturally.
He lowers the cup, swallows, and gives a tiny nose-exhale.
0:06–0:08 — COFFEE TOSS
Mew throws the cup backward using his RIGHT hand.
The cup clears his body completely before spilling.
It must NOT hit his armor, shoulder, head, hair or cloak.
The cup exits frame behind him quickly.
0:08–0:10 — HOLD
Mew’s RIGHT hand is empty.
He remains standing in the same position, still staring forward and slightly upward.
The camera remains natural handheld selfie POV in his LEFT hand.
END with Mew still and ready for the next camera transition.
NO dragon visible.
NO sword draw.
NO phone visible.
NO text, logo, watermark.
Full realism.Pass condition: จบตอน Mew นิ่งแล้วหลังโยนแก้ว ไม่ต้องยัดการโยนกล้องเข้า scene นี้
2.2 Transition A — Camera Launch
หน้าที่ของ shot นี้มีอย่างเดียว: จาก selfie → POV พุ่งขึ้นฟ้า
ถ้า Flow พยายามสร้างมือถือหรือทำ Mew ลอย ให้ใช้ editor ทำ whip 0.15–0.30s แทน
Prompt — Transition A
CAMERA LAUNCH TRANSITION ONLY.
Use the final frame from Scene 1 as the starting image.
WE ARE THE RECORDING VIEWPOINT.
There is NO visible phone or camera body.
Mew remains physically standing on the ground.
Only the VIEWPOINT moves.
The viewpoint receives a sudden strong upward-and-backward impulse and rapidly passes ABOVE Mew’s head toward the space BEHIND him.
The movement feels like a real selfie camera being physically tossed, not a drone, not a digital zoom.
Within the first fraction of a second, Mew leaves the visible frame completely.
After that moment, DO NOT SHOW MEW AGAIN.
End while the viewpoint is airborne and beginning to tumble through foggy sky.
NO dragon.
NO visible phone.
NO airborne Mew.
NO duplicated Mew.2.3 Transition B+C — High Airborne Tumble → Ground Impact
Google Flow settings
- New generation / new shot
- Ratio: 9:16
- ถ้าโหมดนี้ให้ 4s: generate 4s แล้ว trim เอาช่วงดีที่สุด
- Start frame: sky/fog frame จากท้าย Transition A
- ไม่ใส่ Mew reference
- ไม่ใส่ dragon reference
- ไม่ใส่ character sheet
นี่คือเหตุผลที่ช่วงหมุนของเรานิ่งขึ้น: โมเดลไม่มี character ให้เอาไปหมุนกลางอากาศแทนกล้อง
Prompt — B+C Combined
AIRBORNE CAMERA POV — HIGH IN THE SKY, THEN FALLING TO THE GROUND
The camera is ALREADY airborne high above a foggy grass valley.
NO HUMAN.
NO KNIGHT.
NO MEW.
NO DRAGON.
NO CREATURE.
NO VISIBLE PHONE.
NO CAMERA BODY.
WE ARE THE CAMERA VIEWPOINT.
The shot begins high in the sky, already far above the ground.
It must be clear that the camera has already been thrown upward to a great height and is now freely tumbling through the air.
IMPORTANT MOTION LOGIC
This is NOT a single-axis spin.
The camera tumbles in realistic chaotic 3D motion with:
PITCH + YAW + ROLL all happening together.
The axis of rotation constantly changes.
Use changing diagonal axes, irregular wobble, precession, uneven rotational speed, and strong natural motion blur.
DO NOT use:
a simple barrel roll,
one clean end-over-end flip,
one repeated axis,
slow spinning,
smooth floating motion.
TRAJECTORY
At first show mostly grey sky and fog, with only tiny distant glimpses of the ground.
Then gravity starts to win.
The camera clearly begins descending.
The ground appears more often and grows larger over time.
Progression:
already launched high
→ chaotic multi-axis tumble
→ begins descending
→ falling faster
→ grass rushing toward lens.
ENDING
Continue tumbling all the way down.
The viewpoint SLAMS hard into the ground.
Strong physical impact jolt.
One tiny realistic bounce / half-flip is acceptable.
END IMMEDIATELY ON IMPACT.
Do NOT reveal Mew.
Do NOT reveal the dragon.
Do NOT reveal Scene 2 yet.
Full photorealistic live-action physics.2.4 Portrait → Landscape — ทำให้คนดูรู้สึกว่าต้องหมุนมือถือ
Flow clip ที่เริ่ม 9:16 จะไม่กลายเป็นไฟล์ 16:9 จริงกลางไฟล์แบบสะอาด ๆ
วิธีที่เราใช้:
- ช่วง impact ให้ภาพด้านใน จบเอียงประมาณ 90°
- Cut ตอน motion blur หนักสุด
- คลิปถัดไปสร้างเป็น 16:9 จริง
- คนดูจะรู้สึกว่าภาพจาก vlog phone POV เปลี่ยนเป็น cinematic landscape
ตรงนี้ไม่ต้องเสีย generation เพิ่ม ถ้าทำใน editor ได้
2.5 Scene 2 — Cinematic Dragon Reveal
Google Flow settings
- New shot
- Ratio: 16:9
- Duration: 10s
- References: Location + Mew character sheet + Dragon asset + Scene Anchor
- Camera: low, static, cinematic
Prompt — Scene 2
SCENE 2 — FIRST DRAGON REVEAL
Fresh horizontal 16:9 shot.
10 seconds.
One continuous photorealistic shot.
The camera is LOW AND COMPLETELY STATIC in the grass.
It has already landed.
NO tracking.
NO dolly.
NO push-in.
NO zoom.
NO orbit.
COMPOSITION
Foreground: wet grass close to lens.
Middle distance: Mew, 170 cm tall, seen from behind / back-three-quarter, relatively small in frame.
Background: an absolutely COLOSSAL sleeping white dragon.
The dragon feels like a mountain lying in the valley.
Its head alone is enormous compared to Mew.
Its shoulders, folded wings and torso form a huge pale mass disappearing into fog.
STORY STATE
Mew has already thrown away his coffee.
He is standing still.
He has NOT drawn his sword yet.
Sword is fully sheathed on his RIGHT HIP.
Dragon is asleep.
Eyes CLOSED.
Head resting on ground.
Slow heavy breathing.
0:00–0:03.5 — STILLNESS
Mew simply stands still and looks at the colossal sleeping dragon.
He does NOT walk closer.
Wind moves his cloak.
Dragon breathes slowly.
0:03.5–0:05.0 — PREPARE
Mew remains in the same place.
His RIGHT hand moves deliberately toward the sword hilt on his RIGHT HIP.
Sword remains sheathed.
Dragon remains asleep.
0:05.0–0:06.5 — DRAW
Mew draws the sword clearly FROM THE RIGHT-HIP SCABBARD.
The blade physically slides out of the sheath.
A long metallic SCHIIING rings across the field.
Mew lowers the blade into a natural ready position.
He does NOT charge.
0:06.5–0:08.0 — DRAGON HEARS IT
Silence after the metallic sound.
Dragon breathing subtly changes.
Mew remains still.
0:08.0–0:09.0 — EYE OPENS FIRST
The nearest enormous eyelid slowly opens.
Reveal a huge pale amber eye.
The pupil contracts.
Without lifting the head yet, the dragon moves ONLY ITS EYE toward Mew.
0:09.0–0:10.0 — BEGIN TO RISE
Only now does the dragon begin lifting its enormous head from the grass.
Neck starts carrying the weight upward.
Shoulders shift slowly.
Movement is heavy and mass-driven.
END HERE.
NO roar.
NO attack.
NO fire.
NO walking toward dragon.
ABSOLUTE STORY ORDER:
Mew standing still
→ dragon sleeping
→ right hand to right-hip sword
→ sword drawn from RIGHT-HIP scabbard
→ SCHIIING
→ dragon hears sound
→ nearest eye opens
→ eye finds Mew
→ dragon begins lifting head
→ CUT.STEP 3 — Editing / Sound Design
ทำให้คลิปขาย illusion มากขึ้นด้วยเสียงแทนการพยายามให้ AI ทำทุกอย่าง
Scene 1: wind + grass + ice clink + straw sip + cup toss
Camera launch: sharp WHIP
Air tumble: irregular wind flutter / whoomp-whoomp
Ground impact: THUD + grass scrape
Scene 2: quiet field → dragon breathing → sword SCHIIING → low waking rumble
ไม่ต้องใส่เพลงระหว่าง thrown-camera transition ถ้าอยากให้ฟิสิกส์รู้สึกจริง
วิธีประหยัด generation ที่เราเรียนรู้จากการทำจริง
- Still first, video second: แก้ composition ในภาพนิ่งก่อน ถูกกว่าเสีย video generation หลายรอบ
- Asset reuse: ใช้ character sheet / dragon / location ชุดเดิมทุก scene
- One job per shot: ถ้า shot มีเกิน 1–2 งานใหญ่ ให้แยก generation
- Transition ที่ทำใน editor ได้ ไม่ต้องเจน AI: whip, 90° rotate, hard cut, motion blur
- ช่วง airborne อย่าใส่ character reference: ลดโอกาสคนลอย / duplicate character
- อย่า reveal dragon ใน transition: เก็บ surprise ไป Scene 2
- Generate 4s แล้ว trim: สำหรับ transition ใช้แค่ช่วง 0.5–1.5s ที่ดีที่สุด ไม่ต้องใช้ทั้งคลิป
- ใช้ Google Flow โดยตรง: ถ้า workflow นี้ทำได้ใน Flow อยู่แล้ว ไม่จำเป็นต้องจ่าย wrapper เพิ่มเพื่อแค่เข้าถึงโมเดล video เดียวกัน