Per-shot, not vague
One prompt per beat of the 19-second structure — mixing shots is the top cause of mushy renders.
Prompt library
Tiny Desk AI Prompts
Every prompt here was tested against reference-to-video models (Seedance 2.0/2.5, Kling). Copy, attach your photo as @Image1, generate.
- Buy once, generate for 12 months
- No watermark on paid packs
- Renders in about 2 minutes
- 16:9 scenes · crop to 9:16 in one step
The studio the prompts target
The scene art every shot prompt builds toward.
How do you prompt the tiny desk meme?
Describe the set (bookshelf wall, tungsten lamps, small desk), the beat structure (vocal, sax cutaway, band finale), and bind each uploaded photo to a performer with @Image1/@Image2 tags; then lock staging, lighting and timing with a keep-unchanged clause.
How the library is built
Identity tags that hold
@Image1 / @Image2 binding is what Seedance and Kling reference modes are built around — the prompts use it natively.
Model-tested wording
Stage, light and grain clauses are tuned until faces hold through the sax cutaway.
From prompt to post in 3 steps
Step 01
Copy a shot prompt
Pick the beat you want; the prompt already carries the set dressing.
Step 02
Attach your photo
Upload it as @Image1 (and @Image2 for the two-chair cast) in any reference-to-video model.
Step 03
Assemble and post
Splice the beats with FFmpeg, drop the caption, and post the result.
Three findings from testing
From a benchmark pass over the current reference-to-video models on the tiny desk scene.
01
Identity holds at 2, drifts at 5
Two bound photos keep ~consistent faces across all three beats; a five-face cast drifts by the finale. Cast two chairs, imply the rest.
02
Shot prompts beat mega-prompts
One-beat prompts returned cleaner motion and sharper faces than a single do-everything prompt, at equal cost.
03
Set dressing is not optional
Without the bookshelf/lamp/grain clause, models invent bright studios — the meme look lives in that sentence.
The free path: FFmpeg over CapCut
Splice shots, crop 9:16 and burn captions without any editor subscription — three commands, fully local.
Splice the three shots
ffmpeg -i shot1.mp4 -i shot2.mp4 -i shot3.mp4 -filter_complex "[0:v][1:v][2:v]concat=n=3:v=1[v]" -map "[v]" finale.mp4Crop to 9:16 for Reels
ffmpeg -i finale.mp4 -vf "crop=ih*9/16:ih,scale=1080:1920" vertical.mp4Burn the caption
ffmpeg -i vertical.mp4 -vf "drawtext=text='OUTSTANDING':fontcolor=white:fontsize=96:x=(w-tw)/2:y=h-180:box=1:boxcolor=black@0.6" out.mp4The prompt stack
Shot prompts first, then modifiers. Each block copies clean.
Shot 1 & 3 — lead vocal / band finale
A tiny desk concert: the lead singer steps to the vintage condenser microphone at the small cluttered desk and sings the hook full-voice; behind the desk an alto sax and a crimson red keyboard answer each phrase. Packed bookshelf wall, warm tungsten lamps, soft 35mm film grain. @Image1 is the lead singer.
Shot 2 — sax cutaway
A tiny desk concert cutaway: the alto saxophonist takes the solo, fingers on pearl keys, brass glowing under the amber spotlight, the singer softly out of focus at the desk microphone. Bookshelf wall, tungsten lamps, film grain. @Image1 is the saxophonist.
Full band, two photos
The full-band finale at a tiny desk concert: singer at the microphone, saxophonist mid-solo, keyboardist on the crimson keys — all three finishing the hook together in front of the bookshelf wall. @Image1 is the singer, @Image2 the saxophonist; keep every other performer as staged.
Stage & light modifier
Append: packed bookshelf wall, warm tungsten desk lamps, vintage condenser microphone on a stand, crimson red keyboard, deep walnut shadows, soft 35mm film grain, intimate club atmosphere.
Keep-unchanged lock
Append: keep everything else unchanged — staging, framing, lighting, colour grade, and the timing of every beat. Do not copy the face of the performer you replace; use the reference photo. No text, captions, logos or watermarks.
Negative prompt
Extra people, extra limbs, warped instruments, plastic skin, studio whiteout, harsh LED light, text, captions, watermark, logo.
Tiny Desk prompts — FAQ
- Which models do these prompts target?
- Reference-to-video models with @Image binding — Seedance 2.0/2.5 and Kling are the tested pair; the wording ports to Wan and MiniMax reference modes.
- What does @Image1 mean in a prompt?
- It binds an uploaded reference photo to a performer. Whatever you tag @Image1 gets that photo's face; @Image2 binds a second performer.
- Why cast only two faces?
- In our benchmark, two bound photos hold identity across all three beats; five-face casts drift by the finale. Two chairs, implied band.
- Can I use the prompts outside this site?
- Yes — they are plain text. Paste them into any reference-to-video tool; generating here just saves the assembling.
- How do I keep the meme look?
- Never drop the set-dressing clause — bookshelf, tungsten lamps, film grain. It is the single sentence that keeps renders out of sterile studio white.
- Is FFmpeg really enough for editing?
- For this format, yes: concat, crop and drawtext cover splice, vertical and caption. Anything fancier, move to a timeline editor.
More generators & templates
Outstanding AI Video Generator
Cast your face into the 19-second Outstanding tiny desk meme. Lead vocal, full-band finale, red keyboard — one photo, one pack video, about two minutes.
Generate my Outstanding videoAll-Loss Coaches Tiny Desk
Cast your coaching staff into the All-Loss Tiny Desk meme. Five chairs, one headset, zero wins — built from your photos in about two minutes.
Cast my coaching staffUsher "Watch This" Meme
Recreate Usher's "Watch This" tiny desk intro with any face. The point-to-camera beat, the sunglasses, the smirk — one photo, about two minutes.
Make my "Watch This"Ready for your tiny desk debut?
One photo, one scene, one clean clip. The trend moves fast — get your take up tonight.
Run the prompt here