Generating thumbnails

How to generate a thumbnail

The dashboard is a brief, not a prompt box. Filling three fields well beats writing one long paragraph.

Plan Free Cost 1 request (2 images) 6 min read

The dashboard is a brief, not a prompt box

The generation screen is a short form. Each field feeds a different part of the instruction the model receives, which is why filling three of them well beats writing one long paragraph. Top to bottom:

FieldRequiredWhat it controls
FormatYes16:9 thumbnail or 9:16 Story / Short / Reel.
Template nameYesHow the result is filed in Projects. Not visible on the image.
Scene moodOptionalThe overall emotional register — pick a few.
Faceless thumbnailOptionalRemoves the person entirely. See faceless mode.
AvatarOptionalWhich saved face or faces appear. More than one is allowed.
Reference templateOptionalA layout to imitate — from the gallery, your upload, or a video URL.
Thumbnail textOptionalThe headline and secondary line rendered into the image.
Scene descriptionOptionalSubject, setting, key visuals. The highest-value optional field.
Video reactionOptionalThe expression on the face.

Generating, step by step

  1. Pick the format. 16:9 is the default and covers YouTube. 9:16 is a paid format for vertical surfaces.
  2. Name the template. Use the name of the video or the series, not "test 4". This is the label you will filter your library by later.
  3. Set the mood. Two or three moods is usually right. Ten is not — contradictory moods average out into something flat.
  4. Choose your avatar. Select the saved face for this channel. Select several if the video has several people in it.
  5. Add a reference if the layout matters. Skip this when you want the model to compose freely; add it when you already know the shape of the frame you want.
  6. Type the headline. Three to five words, upper case, the payoff of the video. Leave both text fields blank for a clean image you will add text to yourself.
  7. Describe the scene in one sentence. Name the subject, the setting and the light. "A wooden cabin at dusk surrounded by glowing lanterns, fog rolling in" is the right size.
  8. Pick a reaction. Optional, but a face with a defined emotion outperforms a neutral one on almost every video type.
  9. Generate. A progress dialog runs while the two variations render. It normally finishes in under a minute.

You get two variations per generation

One request returns two images and is billed once — the second is free. Do not run the same brief twice to get options; you already have them.

Writing the scene description

This field does more work than any other optional input, and most people leave it empty. It is not a prompt — it is a description of what is physically in the frame. A usable pattern is subject + setting + light + one detail:

  • "A mechanic under the hood of a classic Mustang in a dim garage, one work lamp lighting his hands."
  • "A laptop on a hotel desk at 2am, city lights blurred through the window behind it."
  • "A cast-iron pan mid-flip over a gas flame, flour dust in the air."

Note what is not there: no camera jargon, no style adjectives, no "4k hyper-detailed". The style comes from the mood picker and the reference template. This field is for the content.

After it generates

Both variations land in your library and appear on the dashboard. From there:

  • Open in the editor to change anything — see the editor overview.
  • Download straight away if it is already right. Exports are at the original resolution, no crop.
  • Find it later under Projects, grouped by the template name you set, or in Downloads as a flat grid.

When the result is wrong

Diagnose before you regenerate — a second identical brief usually produces a second version of the same problem.

ProblemFix
Face does not look like youGenerate with a saved avatar rather than a description, and use a clear, front-lit photo.
Composition is close but the person is badly placedDo not regenerate — move them in the editor.
Text is wrong or misspelledFind & replace it in the image, which keeps the original font.
Right subject, wrong feelApply a mood preset in the editor instead of rerunning the brief.
Nothing about it is rightAdd a scene description and a reference template, then regenerate.

Frequently asked questions

Why do I get two images per generation?

One generation renders two variations and is billed once, so you always have something to choose between without paying twice.

Which field matters most?

The scene description. One sentence naming the subject, setting and light source removes most of the randomness, and it is the field people most often leave blank.

Do I have to fill in every field?

No. Only the format and template name are required. Everything else is optional, though a saved avatar and a scene description improve the result dramatically.

How long does a generation take?

Usually under a minute. A progress dialog runs while both variations render.

The face does not look like me. What went wrong?

You almost certainly generated without selecting a saved avatar, or the source photo was heavily filtered. Save a clear, front-lit photo as an avatar and select it in the brief.

Try it on your next video

Sign up free, get starting credits, and make the thumbnail this page just walked you through.

Open Thumblore

Read next

All help topics