English quick take
The safest first video is a subtle head turn, drifting fog, and a slow push-in. Keep the shot to five seconds, keep one person, and keep the reference image as the visual anchor.
Girigo AI Studio 10 min read
Girigo Wish Video
Start from a still portrait, add one controlled camera move, and build a short horror clip without changing the person's identity.
Quick answer
Local clip lab
Trim a short plate and test the grade
Choose a clip you own. The browser trims and grades a preview; no file is uploaded.
No uploadNo accountShort WebM export
Choose a short clip to start. Nothing leaves this browser.
Prompt studio
Start from a structured brief, not a blank box
Pick a look, set the direction, and copy a prompt that names what must stay fixed.
Direction: cinematic portrait; preserve identity; one person.
Use the provided image as the exact visual reference and starting composition. Create a continuous five-second cinematic Halloween portrait animation. Keep the subject's recognizable face, costume, hairstyle, and body proportions consistent with the reference. The subject slowly turns their head toward the camera with a subtle mysterious expression. A gentle breeze moves loose hair strands and the edges of the clothing. Soft fog drifts through the moonlit garden. The camera performs a slow and smooth push-in. Maintain realistic motion, stable anatomy, and consistent lighting. Do not change the person's identity, replace the costume, add characters, or introduce sudden scene cuts. No text or watermarks.
Three useful outcomes
What you can make on this page
Shot plan
A five-second clip with one clear beat
One head turn or one camera move is easier to control than a mini-movie.
Open the studio
Local trim
Test timing before you generate
Use the local video lab to trim a reference clip and preview a filter.
Open the studio
Model routing
Match motion difficulty to the provider
Use budget routes for subtle motion and stronger routes for performance.
Open the studioPrompt lab
Three copy-ready prompt recipes
Five-second portrait
Cursed Wish
A low-cost Wan image-to-video route is a sensible first test.
Use the provided image as the exact visual reference and starting composition. Create a continuous five-second cinematic Halloween portrait animation. Keep the subject's recognizable face, costume, hairstyle, and body proportions consistent with the reference. The subject slowly turns their head toward the camera with a subtle mysterious expression. A gentle breeze moves loose hair strands and the edges of the clothing. Soft fog drifts through the moonlit garden. The camera performs a slow and smooth push-in. Maintain realistic motion, stable anatomy, and consistent lighting. Do not change the person's identity, replace the costume, add characters, or introduce sudden scene cuts. No text or watermarks.
Avoid: identity change, costume change, extra characters, scene cuts, text, watermark
Reflection beat
Mirror Turn
Kling routes are useful when you need a controlled performance beat.
Use the provided still as the exact starting frame. Create a five-second continuous shot of the subject standing in front of a dusty mirror. Keep the face, outfit, and body proportions identical to the reference. The subject slowly lifts their eyes to the mirror; the reflection moves a fraction later than the subject. A cold light flickers once, then settles. The camera holds steady with a very slow lateral drift. Keep realistic motion and stable anatomy. Do not add a second person, do not turn the reflection into a monster, do not cut the scene, and do not add text or watermarks.
Avoid: second person, monster reflection, scene cut, text, watermark
Atmosphere only
Fog Push-In
Fast queues help when you are comparing several motion strengths.
Animate the provided image as a five-second atmospheric horror shot. Keep the subject's identity, pose, costume, and lighting consistent with the reference. Add only subtle motion: slow fog rolling across the floor, a faint pulse in a practical light, and a gentle camera push-in. The subject blinks once and holds a calm, uneasy expression. Maintain realistic skin texture and stable anatomy. Do not change the face, add people, add sudden movements, cut the scene, or add text and watermarks.
Avoid: face change, extra people, sudden movement, scene cut, text, watermark
WaveSpeed, SeeAPI, fal, Replicate, Kie
Which model should run this prompt?
| Provider | Best for | Typical route | Docs |
|---|---|---|---|
| WaveSpeedBudget-first | Low-cost image edits and image-to-video drafts | Seedream edit, FLUX Kontext, Wan image-to-video | Docs |
| KieCharacter-first | Nano Banana image edits and Kling video routes | Nano Banana, Kling video | Docs |
| falSpeed-first | Fast hosted queues and strong model coverage | FLUX, Kling, and other hosted media models | Docs |
| SeeAPIRouting layer | One API surface for several model families and fallback routes | Image and video inference endpoints | Docs |
| ReplicateExperiment-first | Broad catalogue for experiments and batch comparisons | Community and first-party image/video models | Docs |
Step by step
A workflow that protects identity
- Start with a clean still image, not a compressed screenshot.
- Choose one motion beat: head turn, blink, fog drift, or push-in.
- Trim the reference clip locally and check the duration before you generate.
- Use a five-second prompt and keep the camera language simple.
- Compare the first and last frame; if the face drifts, reduce motion strength.
Failure modes
Where most attempts fail
- Do not ask for a costume change, a scene cut, and a camera move in the same five seconds.
- Do not use a low-resolution still and expect a stable face in motion.
- Do not add a second character; the model will often merge identities.
- Do not promise a supernatural result; label the clip as fictional and AI-assisted.
- Do not upload someone else's portrait without permission.
Editorial brief
Girigo Wish Video: the longer brief
A five-second horror clip does not need a plot. It needs one readable change: a head turn, a light flicker, a reflection that arrives late, or fog that crosses the frame. The mistake is asking the model to do a costume change, a camera move, and a scene transition at once. That is how identity drifts.
Generate the still first. Then animate only the still. If the still already has a strong expression, the video model has less work to do. Keep the camera language simple: hold, push in, or drift. Check the first and last frame side by side. If the jawline or eyes move too much, lower the motion strength and try again before you change the model.
Deeper guide
Go deeper before you spend credits
Read the deeper guide
The prompt is only half the job. The field note walks through the failure modes, the identity check, and the model comparison behind this page.
Open the field noteFAQ
Questions this page should answer fast
How long should a Girigo wish video be?
Five seconds is a good first test. It is long enough for one clear beat and short enough to compare models without burning a large budget.
Which model is best for image-to-video?
WaveSpeed's Wan routes are a sensible budget-first test for subtle motion. Kling routes through Kie or WaveSpeed are useful when you need a more controlled performance. fal and Replicate are useful for comparing other hosted models.
Can I edit the video on this page?
Yes. The local video lab trims a clip, applies a preview filter, and exports a short silent WebM in browsers that support MediaRecorder. For MP4 delivery, finish the clip in your editor.
Why does the face change during the video?
The model is inventing motion without enough identity constraints. Reduce the motion, use a sharper reference, keep the shot shorter, and repeat the identity-preservation clause.
Checked sources