Tutor ads can be a screen plus captions, not a twin teacher
Overlay. Ideogram for the snippet. Happy Horse only if you have a consented face.
Overlay. Ideogram for the snippet. Happy Horse only if you have a consented face. A tutor ad that has to sell a lesson is the lesson: an editor, a function, a failing test, a fix. It is not a generated twin of the instructor talking into a camera you never pointed.
Students pause on code. They do not pause on a stranger who looks "like a teacher." Show the screen. Print the hook. Cast a mouth only when the still is a person who agreed to be the ad.
Show the lesson, not a twin
Record the real session. Errors included. Cut the dead air; keep the failure a student will hit on their own machine. A cinematic fake IDE is a product film of a course that does not exist.
The generate aisle is for parts a screen cannot show: a call-stack diagram, a cohort poster. Those are extras. They are not a substitute for the editor.
Do not prompt "a friendly coding tutor explaining recursion at a desk." That sentence casts a person. A person in an ad is a likeness. If you did not shoot them, you do not have them.
Type the hook; generate only a short lockup
The line nobody said — "Week 1 is free," "Watch the test fail," the cohort date — is an authored overlay, not a subtitle and not a generate.
Add a text overlay to a video burns one fixed string for the whole clip. You write caption_text. Position is top, center, or bottom — not a free pixel. Optionally set size, colour, and a background block. The render is an MP4 with that line in the frames. It does not listen. If the tutor actually spoke the line, that is a caption job, not this one. If you needed three different lines at three timestamps, that is timed overlays, not one persistent hook.
Print after picture lock. Overlay before you reframe and you will pay twice: the line that sat clean on 16:9 sits in the chrome after a 9:16 crop.
A short snippet as a still — one signature, a course wordmark, a two-word promise — is Ideogram V4: six credits, text-to-image, text rendering, posters and logos. Quote the string. Proof at 100%. A wrong character is a different function. Aspects: 1:1, 16:9, 9:16, 4:3, 3:4. The text-to-image door is how you run that still.
Ideogram is not an IDE. Asking it for thirty lines of plausible TypeScript is how you ship a screenshot nobody can run. V4 typesets a lockup. It does not compile.
A mouth needs a release
Happy Horse 1.1 Image to Video animates a first-frame image into 1080p with synchronized native audio and multilingual lip-sync. Aspect is inferred from the image. Duration is 3s through 15s. Catalog price is 18 credits. requires_image is true. The still is the speaker.
That is a talking shot off a photograph you own. It is not a way to invent a tutor. If the still is a consented photo of the instructor, write the language and the line, crop to the ship ratio, start at 3s. If the still is a generated "friendly developer," you are running a presenter you cannot stand behind.
A silent Happy Horse brief still tends to come back with sound you did not choose. Prompt whether anyone talks. If the mouth must stay shut, say so.
Capture the real editor. Burn one authored hook on the locked file. Use Ideogram only for a short lockup. Run Happy Horse only off a face you have permission to move — and skip that step whenever you can. The screen already teaches.
FAQ
Can I generate a tutor face if I do not show a name?
No. A nameless synthetic adult is still a person you invented to sell a course. Overlay the lesson. Skip the twin.
Is overlay the same as auto-captions on the lesson audio?
No. Overlay burns a line you wrote. Auto-captions transcribe speech. If the clip has a voiceover you want on screen word-timed, that is the caption path. If the clip is silent code and you need a CTA, that is overlay.
Why not put the snippet in the video prompt so the model draws the code?
Because code is type, and type inside a generate is a typesetting problem. Film the snippet, or lock a short string in Ideogram and composite it. Do not let a video model invent a function.
What if I have a photo of myself, the tutor?
Then Happy Horse is in play: that photo is frame one, you write the line, you start short. The release is yourself. That is the only case where a talking twin is not a twin.