来自 X · Motion Transfer
动作迁移提示词
从 X 公开帖子精选的 Motion Control / Motion Transfer 视频提示词，可在 Oraclip 试用，并保留作者署名与原帖链接。
Did the motion control test with @higgsfield new native model. Genjutsu.
Did the motion control test with @higgsfield new native model. Genjutsu. This model excels in staying true to the original motion, even better than seedance 2.5 edit. Although, while it excels in motion, it can lack with surroundings details if you don't specify your prompt. https://t.co/xMMGHu7qLB
Gemini Omni's motion control is f*cking cracked i just figured out how t
Gemini Omni's motion control is f*cking cracked i just figured out how to turn 1 reference video into 50+ AI videos with the exact same movements... you have a video of someone eating, dancing, using a product, doing whatever complex motion you need. you feed it to Gemini Omni and it recreates that exact motion with a completely new AI character in literally one prompt i've tested this against Kling motion control and it's not even close. Kling falls apart the moment you try anything complex. eating scenes look weird, hand movements get mangled, anything multi-step breaks down completely. Gemini Omni handles all of it if you're still using kling motion control or paying creators to split test your videos, this replaces that entire workflow here's the thing though. you can't just prompt this out of the box. if you try to do motion transfer with default prompting you're going to get errors or the motion won't transfer properly. there's a specific prompting method that makes it work every time so i packaged up the whole system.. here's what you're getting: > full step by step video breakdown > how to find the best reference videos to use > the exact prompting system that allows for motion control transfer so you never get errors > the workflow for batching this out at scale (1 video → 50+) RT + reply "MOTION" and i'll send it over (must follow so i can dm)
It's crazy how easy AI video editing has become. 🤯 This was made using M
It's crazy how easy AI video editing has become. 🤯 This was made using Mimic Motion in the @itsPolloAI app. I simply used a motion reference video together with another image, and the app recreated the same movements, all directly on my phone! No complicated editing or desktop software required. Just a few taps and the result is ready.
🔥 How to Create Dancing AI Influencer Videos in Minutes! You can make yo
🔥 How to Create Dancing AI Influencer Videos in Minutes! You can make your AI influencer replicate dance moves and facial expressions from existing videos. Here’s how 👇 1. Create Your AI Influencer In OpenArt, select Nano Banana Pro, add your image prompt, choose 9:16 or 16:9, set resolution to 2K, and generate. 2. Choose a Dance Video Download a dance video from Pinterest, TikTok, or Instagram, then screenshot its first frame. 3. Match the Starting Pose Upload your AI influencer image and the screenshot to OpenArt as visual references. Generate an image matching the pose. This step is optional but can improve results. 4. Animate the Dance Open OpenArt’s Motion Sync tool: → Select Kling 2.6 → Add your influencer image as the subject → Upload the dance video as the motion reference → Set Character Orientation to Exact → Leave Video Mode on Standard → Enable Keep Original Sound if needed → Click Generate That’s it! Your AI influencer can now recreate the dance movements and expressions.
Depth video workflows have been getting a lot of attention lately, so I
Depth video workflows have been getting a lot of attention lately, so I tested one myself. Combined with Seedance 2.0, it produced more natural motion-transfer results than using Kling Motion Control directly. Why use a depth video? 1. It removes the original character and scene details, reducing copyright and sensitive-content risks. 2. It preserves the original motion, timing, and spatial structure. This separates motion extraction from visual generation, allowing you to recreate the movement with better models and any reference character. We’ve also launched a free online tool that converts regular videos into depth videos—no local setup required: https://t.co/mLq2yAYTyx In the example below, we converted a dance video from Douyin into a depth video, then regenerated it with a reference character using Seedance 2.0. The original choreography and timing are preserved, while the lighting adapts naturally to the new character and scene.
I made a bird-whisperer still in Adobe Firefly Boards with GPT Image 2.5
I made a bird-whisperer still in Adobe Firefly Boards with GPT Image 2.5 Flare, then turned it into an 8-second video clip with Veo 3.1 without leaving the board. Which one looks better? Vertical or widescreen? Workflow I used: New Board Generate image → Flare → landscape stills Place both on the canvas Generate video → Veo 3.1 First frame + last frame from the canvas Hit Generate and wait (400 credits, 8s, 720p) Different models ask for different things. Veo wants first/last frame. Firefly Video can ask for a motion-reference video. Kling Omni wants shots. Once you know that, Boards stops feeling confusing. This is the kind of use case I wanted to show: still → motion in one workspace. Check the comments section for screenshots of the prompts used and the screen recordings on how I created both images in landscape and the vertical short video clip on Adobe Firefly Board. I also explain why I ended up generating the video clip in widescreen.
How to use Kling 3.0 Motion Control for football-style shots? Prompt bel
How to use Kling 3.0 Motion Control for football-style shots? Prompt below! Who else lives for this kit? Tested a Brazil football portrait with Kling 3.0 Motion Control and the result came out super clean. Started from a full-body stadium portrait of a blonde woman wearing the Brazil home kit — yellow jersey, black shorts, black socks, white sneakers — then used motion control to turn it into a more dynamic football-style shot with that cinematic stadium energy. 1. Go to the [Kling AI Video Generator](https://t.co/jz38n6Kydy) 2. Write your full prompt or add reference images 3. Upload the image you want to animate 4. Click Generate and get your animated video # Photo prompt: "A full-body portrait of a young woman with long straight blonde hair, fair skin, and a slight smile, standing confidently in a football stadium. She is wearing the official Brazil national football team home uniform: bright yellow jersey with green collar and sleeve cuffs, CBF logo on the chest, black shorts, black socks pulled up, and white football sneakers. Her pose is casual with one leg slightly forward, arms relaxed. Background shows a vibrant football stadium with green pitch, crowded stands, bright stadium lights, and athletic atmosphere. Realistic photography style, sharp details, natural daylight mixed with stadium lighting, high resolution, cinematic composition." The base image was designed with: * full-body composition * realistic football stadium background * bright match-day atmosphere * natural pose and confident stance * sharp photoreal details Really liking how Motion Control helps keep the composition stable while adding smooth camera movement. Share your thoughts about this dance video for the World Cup 2026 below!
Impressed by GPT-Astra’s CoT ! This Kung Fu demo is built from scrach in
Impressed by GPT-Astra’s CoT ! This Kung Fu demo is built from scrach in Blender and takes the reference video + simple prompt as input. The CoT reasoning really show a grasp of kinematics, joint coordination, and body motion ! Reference video comes from PhysAvater @qingqing_zhao_ @yang_zheng18
this use case for Seedance 2.5 is insane... someone turned a walkthrough
this use case for Seedance 2.5 is insane... someone turned a walkthrough of a half-built construction site into a finished, furnished marketing video the trick: Seedance 2.5's video-to-video mode. feed it a driving video and it re-skins everything while keeping the motion here's the setup: - raw walkthrough filmed on the unfinished site = driving video - 2 generated stills of the finished space = start and end anchor frames - camera path, pacing, doors, turns... all pulled straight from the real footage - building, finishes, people in shot... all pulled from the anchor frames prompt structure: "Use [reference video] as driving video. Start with [reference image]. POV, camera follows a real estate agent walking towards a newly completed apartment. He's showing off the place to camera, walking casually up a footpath surrounded by lush landscaping. Camera follows him through the glass doors into a furnished common area, marble floor, oak timber wall panelling. Ends with [reference image]. No transformation, no morphing, no bgm." that last line is the whole trick. without it, Seedance drifts off your camera path trying to be clever with it, motion stays locked to what you actually filmed where it breaks: material detail it never saw, marble veining, panel grain, reflections. fine for a teaser, not a client-facing final for real estate and architecture marketing this changes the math. you don't need the building finished to shoot the video that sells it, you need 30 seconds of someone walking the path and Seedance 2.5 does up to 30 seconds in one pass, so that's an actual walkthrough shot, not 6 clips stitched together
This is f*cking gold, and most people will scroll right past it 🔥 While
This is f*cking gold, and most people will scroll right past it 🔥 While everyone chases one viral AI clip, I built a character who pulled 27.7K followers in 30 posts across Instagram, TikTok and Facebook. Biscuit Johnson remains true to himself. That’s what makes his style unique. My last post about him hit 230K+ views. This 360° turn is the real secret: one face, locked from every angle, so he looks the same in every video. The exact workflow: -> Give him 3 loud traits. Bald top with wild bushy sides, thick mustache with a gap tooth, tan double-breasted suit over a black turtleneck. One outfit forever. -> Generate the face in Nano Banana Pro (Google Flow), then a character sheet: front, side, back, close-ups -> Turn the sheet into a 360° video like this one. Every clip pulls from it -> Screenshot a frame from a 1M+ clip, generate Biscuit in that angle, run both through Kling or Higgsfield -> Post 1-2 Reels a day. “Certified Menace” in the bio, no text on the video -> Route traffic: bio → Linktree → your offer A viral clip gets watched once. A character gets followed 👀 Prompt 1: 360° turn (Kling / Higgsfield / Seedance) Upload your character sheet as the reference, not a single photo. The back view matters. Prompt: The same man from the reference stands on a plain light-grey studio background and slowly turns 360 degrees in place, clockwise, at a steady speed: front, side, back, side, front. The person rotates, the camera does not move. Full body in frame, arms relaxed, neutral expression, feet stay on the same spot. Soft even studio light. Same face, hair, outfit and proportions in every frame. Prompt 2: motion swap (Kling Motion Control / Higgsfield / Seedance) Inputs: your character image + a viral video with one person and a steady camera. Prompt: Replace the main person in the video with the character from the image. Keep the same face, hair, outfit and body proportions in every frame. Copy the original movement, gestures, expressions and timing exactly. Keep the original background, lighting, camera movement and audio. Lip sync if the person talks. Match the source video quality, no glossy AI look.
@higgsfield_ai Genjutsu is wild. 🤯 Genjutsu allows you to upload a motio
@higgsfield_ai Genjutsu is wild. 🤯 Genjutsu allows you to upload a motion reference and transfer the movement, acting, and overall performance to your own character. For this test, I used an existing video as my motion reference and applied the movement to my own avatar. The https://t.co/3PLwktCouP
THIS AI KUNG FU MASTER HAS ALREADY EARNED $7,657 IN JUST TWO WEEKS. HE H
THIS AI KUNG FU MASTER HAS ALREADY EARNED $7,657 IN JUST TWO WEEKS. HE HAS 34,600 FOLLOWERS AND HE DOESN'T EVEN EXIST. YOU CAN START MAKING MONEY FROM THIS RIGHT NOW, JUST LIKE HIM. here’s how to build your own character and make videos like this ↓ — meet Sean Dynasty Prosper. 26 posts. 34.6K followers. an AI-generated profile built around a “30th Generation Wing Chun Master.” dramatic curved hair. dark clothing. bright white cuffs. in this clip, the characters take turns approaching the camera before performing together. one recognisable cast gives the creator a whole series of situations to test. — 1 → CREATE YOUR CHARACTER use ChatGPT image generation or Higgsfield’s image tools to design an original character. give them distinctive features that stay readable during movement: hairstyle, silhouette, outfit and facial expression. save: → a clear facial close-up → a full-body image → front, three-quarter and side views → consistent clothing and hair keep the reference sheet for future scenes. for animation, prepare a clean individual image of the character in the setting you want. — 2 → PREPARE A MOVEMENT REFERENCE record a short dance or playful martial-arts sequence, or use footage you have permission to work with. start with one person whose hands and feet remain visible. choose a short action with a clear beginning and ending. you’ll be able to inspect the result and retry without regenerating a long performance. — 3 → PICK YOUR VIDEO WORKFLOW Kling Motion Control → animate your character using a performance reference upload your character image and the movement video. the reference video guides the actions. facial Element Binding can help preserve the character’s identity through movement and head turns. Kling’s guide supports one bound character, so begin with solo shots before attempting a whole crew. Higgsfield Genjutsu → transform existing footage choose Motion Transfer to apply a reference performance to a new character or scene. choose Object Swap to replace selected elements while retaining the rest of the shot. use the workflow that matches your goal: rebuilding the scene around a movement, or changing part of existing footage. — 4 → ASSEMBLE THE VIDEO IN CAPCUT pick the strongest takes. start with the action. remove empty pauses. add suitable audio. export vertically. for the first test, give each character a separate shot. once those work, try bringing them together. — 5 → CHECK THE DETAILS watch the result closely: → does the face change during turns? → do the hair and clothes stay consistent? → do hands distort or disappear? → do feet slide across the ground? → do characters merge when they overlap? regenerate weak shots. trimming helps, but it won’t fix every visual error. — THE STARTER SETUP character references → ChatGPT or Higgsfield movement transfer → Kling Motion Control footage transformation → Higgsfield Genjutsu final edit → CapCut choose one video workflow first. you don’t need subscriptions to every service to begin testing. start with one character and one short performance. build the crew once the first result holds up.
1/ Quick workflow for this case: 1. Take the original reference video 2.
1/ Quick workflow for this case: 1. Take the original reference video 2. Convert it into a Depth video 3. Use the Depth video as the motion reference 4. Add your own character reference image 5. Use a prompt to replace the outfits and styling 6. Generate with Seedance Original video reference: Douyin creator 原来是陶阿狗君 Full Depth workflow is in the quoted post above.
ロボットダンス... 再挑戦‼ フリーの素材と Kling 3.0 でやってみたよっ ちょっとそれらしくなったかなぁ🤔 #Kling #AI美女
ロボットダンス... 再挑戦‼ フリーの素材と Kling 3.0 でやってみたよっ ちょっとそれらしくなったかなぁ🤔 #Kling #AI美女 Kling 3.0 の Motion Control なかなか凄いな👀 フリーじゃない素材でやってみたら、もっといい感じになったけど、公開できない💦 Kling 内の素材で作ったのは公開していいんだろうか、ちと調べるかな アイソレが好きなんだけどなぁ
The best motion control is on KLING AI. Referral code in replies. https:
The best motion control is on KLING AI. Referral code in replies. https://t.co/m14Jr8sfqD
🔥 This looks like a scene from an AI movie… But the movement came from a
🔥 This looks like a scene from an AI movie… But the movement came from a real reference video. Using @pixverse Depth Map Control, I combined: 🎬 Real motion reference 🌐 Depth guidance 🖼️ A new visual style The result? A character keeping the original action while stepping into a completely different world. From fight choreography to cinematic scene transformation, this workflow opens up new creative possibilities for AI video creators. More direction. More control. More ways to create. 👀 https://t.co/fVrLEdnw4j
Randi, allow me to demonstrate. 😂💃 I borrowed your video, dropped myself
Randi, allow me to demonstrate. 😂💃 I borrowed your video, dropped myself into it with Kling motion control and… yeah. This is why I keep recommending it. 😏 Same moves. Different girl. ❤️ Your turn, @RandiAthenas — give Kling another shot! 😂 #MikaMilani #KlingAI #MotionControl #AIVideo
A 21-YEAR-OLD GUY MAKES $10,000/MONTH WITHOUT FILMING OR ACTORS While br
A 21-YEAR-OLD GUY MAKES $10,000/MONTH WITHOUT FILMING OR ACTORS While brands spend thousands of dollars renting studios, finding models, and editing, this guy creates ready-made UGC advertising content in 3 minutes directly from his phone. No face in the frame, no expensive equipment, and no burnout. What's the gist of the process? He uses a combination of character generation and Motion Control (AI-powered motion transfer). Here's the whole process step by step: Photo model generation: Using Higgsfield AI, he creates or uploads an ultra-realistic model (indistinguishable from a real person, just from the outside). Motion recording: The guy simply grimaces in front of his phone camera—making the desired facial expressions, gestures, and delivery. The magic of Motion Control: He uploads his motion video to the Kling Motion Control module and links it to the generated AI model. Result: The neural network will precisely replicate every gesture and look on the face of a perfect AI model! How to make money from this? Selling UGC advertising to brands: Agencies and stores pay $100 to $500 for a single short video for TikTok Ads or Reels (or you know how else this can be used on other platforms). Creating AI influencers: Launching a virtual blogger who can advertise products 24/7. Traffic arbitrage: Uploading dozens of unique videos a day to offers without being banned for duplicate content. Bookmark this to replicate it and finally start earning real money.
Tried the same choreography-driven animation on Kling back in February.
Tried the same choreography-driven animation on Kling back in February. Redid it with Seedance 2.0 this week. Motion fidelity is a clear step up. This is currently my go-to for motion control. https://t.co/cxkIJEkYQ4 https://t.co/nA1SYwVW3S https://t.co/BrhMWVHfWd
You guys don't know how powerful ghostfeed mcp is There are three steps
You guys don't know how powerful ghostfeed mcp is There are three steps to producing a video of this quality: 1. Enter the tiktok link in claude once ghostfeed mcp is installed 2. Ghostfeed analyses the original clip and writes high quality prompt is exact and creates first frame which is exact. 3. Don’t use Kling Motion Control. Use Seedance Mini
Finally, i can sleep well tonight. Higgsfield genjutsu is perfect motion
Finally, i can sleep well tonight. Higgsfield genjutsu is perfect motion control for cardistry. . I pushed it beyond simple motion transfer: • Copied the original motion • Changed the card colors • Replaced the characters • Added minimal motion-graphic typography • Kept the visual timing and choreography intact And surprisingly, it executed the whole concept really well. . made with @higgsfield_ai @higgsfield_crea
£142,485.84 FROM AN AI GIRL BUILT ON 5 TOOLS AND 1 NICHE. Here's the sta
£142,485.84 FROM AN AI GIRL BUILT ON 5 TOOLS AND 1 NICHE. Here's the stack. Claude writes every caption, every reply, and remembers each fan. Nano Banana Pro blends 2 real faces into a third that belongs to nobody. Kling Motion Control takes a viral clip and transfers the movement onto her, frame by frame. ComfyUI and Wan2.2 on RunPod handle the higher end renders. CapCut runs 1 pass that turns the render into a phone photo. Dark hair, a small scar above one brow, alt streetwear. She skips class on Tuesdays, posts a birthday dinner with fake friends, texts fans first when she's having a bad week. No camera. No studio. No second person on payroll. A pretty face gets 1 scroll. A girl with a niche gets followed for years. U just watched a business plan, not a girl.
Too funny not to post. First attempts working with local video gen with
Too funny not to post. First attempts working with local video gen with motion control net. Ways to go, but it's a start. Vesper x Apex dance scene w/ audio https://t.co/zl1WU4JoqM
40 MINUTES IS ALL IT TAKES TO BUILD THE GIRL DANCING IN THIS REAL APARTM
40 MINUTES IS ALL IT TAKES TO BUILD THE GIRL DANCING IN THIS REAL APARTMENT. SHE MOVES LIKE A HUMAN FOR ONE REASON: A REAL HUMAN MOVED FIRST. HERE IS THE BUILD. Most people assume the realism comes from the model. It comes from a person you never see. The build: 1. Train a LoRA on 15 to 30 reference images. It locks her face, jaw, and exact skin tone so she never drifts across outfits or angles. 2. Film a real dancer doing the movement. Ring light, empty apartment, a dog asleep in the corner, all real. That footage is only the skeleton she will borrow. 3. Kling Motion Control maps the dancer's joints onto her body, frame by frame. Hair swing, fabric, weight shifting on the hips, all inherited from a human who was actually in that room. 4. Prompt the outfit and lock it to one palette. Pink sweater, pink bag, pink boots, nothing clashes because the color script is set once and reused. 5. Upscale and run a skin pass. Pores, texture, a slight sheen under the ceiling light, so she survives a full-screen zoom on a phone. 6. Generate the caption and voice from her persona file, so every post reads as one consistent girl living one consistent life, not a fresh render each time. Once the LoRA is trained, the whole clip takes 30 to 45 minutes. New plate footage in, new clip out. Here is the part that should change how you watch these. That uncanny smoothness you cannot place is not the AI being brilliant. It is a real dancer's body, quietly wearing a face that was never in the room. The apartment is real. The dog is real. She is the only thing that was never there. How the full motion-transfer pipeline is built is in the article below.