Course slides are Qwen/Ideogram, then I2V if needed

The slide is the product. A photoreal hero and a course frame are both "images." They are not the same job. A hero is judged on light. A slide is judged on whether every word is the word you specified, at a size a student can read on a phone. Send a curriculum still to a portrait model and you get text-shaped noise.

Lock the page on an image row. Motion is optional, and it is last. Do not image-to-video a typo.

Route the still by how much type it carries

Body copy, columns, a numbered list, a source line. That is a page. Qwen Image 3 Pro when the slide is the product is the routing rule. The published family page is Qwen Image 3 Text to Image: the 4-credit sibling is the layout scout; the Pro tier is 8 credits, up to 2K, text rendering listed. Scout cheap. Keep on Pro.

Eight to twelve words that must read at thumbnail size. Title card, module stamp, "Lesson 04." That is Ideogram V4: 6 credits, posters and logos listed. Quote the string. Do not ask Flux to spell the course name. Ideogram owns words in the frame is the lettering split.

A photograph of a desk with a slide in it. Generate the slide, then composite. Do not put readable body copy in a lifestyle scene.

Text-to-image is the door. Quote every string you will proof. "Week 3: the 12-bar blues in G" is a string. "A headline about week three" is how you get a plausible headline you cannot ship. If the type will be overlaid later in the slideshow maker, generate the plate empty.

Proof the slide before you buy motion

Students zoom. Thumbnail impression is not the test. Proof at 100%. One wrong interval name is a regenerate.

Only then decide whether the still needs to move. Most course ads do not. A locked title card plus a voiceover is a still and a TTS row. A Ken Burns push is image-to-video off the approved file, with a prompt that describes only the camera. Do not re-describe the bullets in the video prompt. Motion will not correct a misspelled chord.

Do not prompt a video model to hold "MODULE 07 — MIXOLYDIAN" for eight seconds. That is typesetting across two hundred frames. Generate the card. Overlay if the string must be exact. Animate only if the card is already the product.

The course audience is the lesson, not the thumbnail

Online course creators convert with a real lesson excerpt, not a testimonial reel. A generated slide that cannot be read is not that excerpt. A talking synthetic lecturer is a different catalog row and a likeness you probably do not have.

Use one real frame from the curriculum: 16:9 for the YouTube lesson and a 9:16 crop of the same still for Reels. Do not generate a second curriculum for the vertical. If the type dies in the crop, the slide had no safe zone.

Three posts a week is the cadence on that audience page. That is a still pipeline with a proofing loop, not a weekly film.

FAQ

Should every slide be Qwen Image 3 Pro?

No. Scout on the 4-credit model. Keep on Pro when someone will proof body copy. Title cards with a handful of words are Ideogram.

When is image-to-video actually needed?

When the approved still is the ad and the brief wants a push or a first-to-last frame between two signed-off cards. It is not needed to "see if the type holds."

Why not overlay every string and skip these models?

Do that when you already have the lockup. Use Qwen when the slide is the image. Use Ideogram when the lettering is the illustration.

Will a free lesson hurt sales if the slide is the ad?

The online courses page's answer is no — if the lesson is actually readable. A garbled generate is not a free lesson.