Turn yourself into your own drawing with ChatGPT and Higgsfield Genjutsu
Hey, it's Cindy 🌱 You commented PLUGIN, so here's the whole thing from my video: how I drew my own character, the full Genjutsu prompt that maps it onto my body following my every move.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
read this first
Pick the right mode: Motion transfer
Higgsfield Genjutsu has two modes, and they do very different jobs. Turning yourself into a drawing is a Motion transfer job: it keeps your exact movements, camera and timing and rebuilds you as your character. Pick Motion transfer before you generate, and the prompt names it too so the plugin lands on the right one.
Motion transfer ✅ use this Keeps your movement, camera and timing, and rebuilds you as your drawn character following every move. This is the one that turns you into the drawing.
Objects swap Changes one element and keeps the rest of your real footage, like swapping a single person or object in a scene you want left untouched. A different job, not this one.
💡 Your drawn look comes from the reference images you upload, and your movement comes from your video. The prompt just names Motion transfer and locks the look and timing. Higgsfield's Genjutsu page describes both modes.
step one
Draw your character
You do not need to be an illustrator. Work in any layered drawing app, like Procreate or Photoshop, or a free one like Krita on desktop or Ibis Paint X on your phone. Layers are the only must-have, because they let you keep your guide, your lines and your colours separate.
🔴 Block the rough structure in red. Drop a frame from your original video in as a reference on the bottom layer. On a new layer above it, rough out the pose and proportions in red. Red keeps these guide lines obviously separate from everything you draw on top.
✏️ Line art on its own layer. Make another layer and trace the line art over your red guide. It can be pretty rough, the colour and the render clean it up anyway.
🎨 Base colours, then highlights, then final touches. Under the line-art layer, drop in flat base colours. Then add your hair highlights and the small final touches, like blush, a highlight in the eyes, or a shine on the hair.
🧍 Export it framed like your video. Turn off the red guide layer and export just the character. Match the framing to your video: filmed waist-up means a waist-up character, with hands visible if your hands are in the shot. This is the file you hand to ChatGPT as Image 1.
🎨 Clean, flat, front-facing art comes out best. A busy background behind your character, heavy texture or a three-quarter angle gives Genjutsu more to misread. A plain or transparent background is ideal.
step two
Prep your video and your character
Genjutsu rebuilds the person frame by frame, so the cleaner your inputs, the less it has to guess. These make the biggest difference.
📱 One person, one shot, steady camera. Put your phone on a tripod or lean it on something. No cuts, no zooms, and nobody else walking through. Keep it under 30 seconds; 10 to 15 seconds holds best.
💡 Film in even, bright light. Window light facing you or a ring light. Dark rooms and motion blur are where the render falls apart.
🖐️ Keep your movements natural, not fast. Talking, turning your head and gesturing are all fine. Very fast hand movements are where extra fingers and smears show up. Plan one clear hand movement for the CapCut cut later.
🧍 Match the character to the video's framing. Waist-up video means a waist-up character, full body means full body. Mismatched framing is one of the main causes of warping.
🙂 Front-facing character, hands visible if yours are. A character with a build roughly similar to yours, and hands drawn in if your hands are in the shot, comes out far more cleanly.
step three
Set it up in ChatGPT
🔌 Connect the Higgsfield plugin. In ChatGPT, desktop or phone, open Plugins, search for Higgsfield and hit Connect. You'll need a Higgsfield account with credits.
🖼️ Start a new chat and upload your files. Add your video and your character art. Two to four images of the character at slightly different angles hold up far better across a full clip than a single one.
✍️ Type @higgsfield /genjutsu, then paste the prompt below. Fill in the one choice in brackets (keep the character's outfit, or match the one you wear in the video).
🔁 Check the test render, then go to 1080p. The prompt makes ChatGPT run a cheap low-resolution test first, check it for problems and fix them before you spend credits on the final.
copy this
The full prompt
🎨 Turn me into my drawn character
I've attached my video (the source) and images of my drawn character. Use Higgsfield Genjutsu in Motion transfer mode: keep my exact movement, camera and timing from the source video, and rebuild me as the drawn character from the images.
Before generating, check the files and tell me in three short bullet points if anything will hurt the result: the video should be one person in one continuous shot with a steady camera, 4 to 30 seconds; the character images should show the same character with a clear, front-facing face, framed like the video, with hands visible if my hands are visible; and the character's build should be roughly similar to mine. If something is off, ask me whether to continue.
Generate a 480p test first with these locks:
Look: the character's face, hair, outfit and hand-drawn style come 100% from the character images, held consistent in every frame, including in profile, mid-blink and mid-word. Outfit: [keep the character's outfit / match the outfit I wear in the video]. Keep clean line art and flat, soft shading the whole way through, with no drift between frames.
Movement: copy my performance from the source video frame for frame. Every head turn, blink, glance, hand gesture, shoulder movement and pause happens at the same moment, at the same size and speed. Mouth shapes follow my speech so the lip sync lines up with the original audio. Hands keep five fingers and follow my hand positions.
Camera and timing: keep the source video's framing, crop, camera movement, length and frame rate exactly.
Avoid: the face or body morphing between frames, the hand-drawn look flickering or turning photo-real, extra or missing fingers, smeared hands, lip movement out of time with the audio, and any new zooms, camera moves or cuts.
After the test, tell me what failed: the look holds every frame, movement and lip sync in time, clean edges and hands. For anything off, add one line saying exactly what went wrong and how it should look, then run again. If the face or hands drift, add more reference images of the character at different angles. Once it holds, run the same prompt at 1080p.
In Motion transfer your reference images do the heavy lifting, so the prompt stays short: it names the mode, locks the hand-drawn look to your images, and ties the movement and timing to your video. More character images at different angles beat a longer prompt every time.
step four
Stitch the reveal in CapCut
The clean flip from the real you to your drawing is a cut hidden on a hand movement. Both clips show the same motion, so the cut is invisible.
📥 Import both clips. Bring your original video and the Genjutsu output into CapCut.
🎞️ Stack and line them up. Put them on two tracks and slide them until the same moment sits on the same frame. Matching the audio waveforms is the easiest way to line them up exactly.
✋ Cut both on the hand-open. Find the frame where your hand opens, split both clips there, then keep the real video before the cut and the Genjutsu video after it.
📤 Export. You get one continuous clip that flips from you to your drawing right as your hand opens.
💡 Any clear, fast gesture works as the hiding point: a hand passing the lens, a quick turn, a snap. The faster the motion at the cut, the less the eye can catch it.
if it's still not right
Fix the usual problems
🏠 The background looks off Motion transfer rebuilds the whole scene, so the drawn half has an AI background by design. To steer it, add a reference image of the background you want, or keep it plain. Your real room only needs to show in the real half before the CapCut cut.
🎨 The drawing loses its look Add more reference images of your character at different angles, keep the "hand-drawn look" line in the prompt, and export your character cleaner, flat colours and a plain background, so Genjutsu has less to reinterpret.
🖐️ Hands or hair look wrong Draw the character's hands in if your hands are in the shot, keep its hair roughly your length, and slow down any quick gestures when you film.
👄 Lip sync drifts Face the camera for the talking parts, keep your face unobstructed, and split a long video into shorter clips you render one at a time.
what it costs
Credits and limits
480p test 40 credits for a 15 second clip. Use this for every first attempt.
720p 104 credits for a 15 second clip.
1080p final 144 credits for a 15 second clip. The cost shows before you generate.
Clip length Videos up to 30 seconds, output up to 1080p.
Only use a face you have the right to use: your own, a character you drew, or someone who has said yes. Higgsfield's terms of use require consent from every person whose likeness appears, and you need to disclose AI-generated content wherever the platform or the law asks for it.
🤝 Higgsfield partnered with me on this video, and the Higgsfield links on this page are affiliate links.