A repeatable short-form pipeline.
Veo → Runway → Resolve. The cuts, the prompts, the project file.
This is the pipeline for shipping five short-form videos per week solo. Five tools, one locked 15-second template, five production steps.
What you'll have when you finish: a saved 15-second beat structure, a Veo prompt scaffold filled with your subject/style axes, a pinned ElevenLabs voice config locked at stability 0.5, a Resolve export preset at 1080×1920 / 30fps / 12 Mbps, and a four-block production routine that turns ideas into shipped videos without remaking decisions every time.
Accounts you'll need: labs.google/fx/tools/flow (Veo) · runwayml.com · elevenlabs.io · DaVinci Resolve (free) · Topaz Video AI.
The stack — five tools, ranked.
- 01Veo 3 — hero shotsweekly
- 02Runway Gen-4 — motion & editsweekly
- 03Higgsfield — camera movesmonthly
- 04ElevenLabs — voice & SFXweekly
- 05Topaz Video AI — upscalemonthly
Veo makes the shot. Runway moves it. Higgsfield handles cinematic camera. ElevenLabs is the voice across everything. Topaz is the polish before the export.
How to apply it.
-
01template
Lock the template — once, then never again.
15 seconds, three beats. Open Resolve, create a project at 1080×1920, 30fps, 16-bit. Save as
Template — Shorts v1.drp. Set defaults you'll never touch again:— Color grade: one LUT applied at the project level (not per clip)
— Caption font + size: locked across the project
— Outro card: 2-second clip with your brand mark, imported as a media pool item
— Click track at 90 BPM in the audio bin (cuts land on the beat)The template is invisible to viewers and load-bearing for you. It's why you can ship five pieces this week and one took you all month last quarter.
-
02hero · 90 min
Hero shot in Veo — three rolls, pick one.
Fill in the 6-axis prompt scaffold (in the Build-along tab): subject, action, environment, camera, light, style, plus a negative list. Generate three variants per piece. Pick one. Never re-roll past three.
If no shot lands in three rolls, the prompt is wrong, not the model. Rewrite one axis at a time — change the lens, then the lighting, then the time of day. Don't grind the same prompt.
Save the chosen file to your project's
01-hero/folder with the prompt as the filename so you can re-find it later. -
03motion · 2 hrs
Motion & edits in Runway.
Batch all five pieces in one Runway session. Same camera moves, same transition style, same color profile. The session is where the consistency lives.
Cuts are on the beat. Pre-define the beat with a click track. Random cuts are the giveaway that the piece was made in a hurry.
-
04voice · 60 min
Voice over in ElevenLabs.
Single voice across the campaign. Fixed pace, fixed stability setting (lock it at 0.5). Mix audio at -16 LUFS for social, -23 for ads.
Never switch voice models mid-campaign. The viewer hears it before they see it.
-
05ship
Upscale, export, queue.
Topaz batch on all five files with the same model and the same artifact threshold. Export to 1080×1920 H.264 at 12 Mbps. Queue all five posts at once. Walk away.
What we stopped doing.
- ×Reinventing the format every week. Format consistency is what reads as a brand.
- ×Mixing voice models in a campaign. Ears notice faster than eyes.
- ×Editing in different software for "look variation." The look comes from the grade, not the editor.
- ×Posting whenever. Same time, same day. Algorithms reward consistency.
- ×Rolling a hero shot ten times. Three rolls. Pick one. Move on.
- ×Custom transitions per piece. One transition family per campaign. Save the rest for the next campaign.
The take.
Output isn't about better tools. It's about fewer decisions made each week. Lock the template once. Vary the content. The pipeline's job is to be boring — that's what makes it fast.
Steal one thing: the three-roll rule on hero shots. If the model can't land it in three, the prompt is wrong. Rewriting prompts is faster than rolling.
Don't add these until the template has shipped through four campaigns. They compound when the basics are boring.
Template branches.
Three template variants: hero (15s, generative-heavy), talking-head (20s, mostly face), B-roll (12s, montage). Same color grade, same caption style, same outro across all three. Branches let you serve different feeds without abandoning the system.
Pre-generated B-roll library.
Once a quarter, run a 50-shot Veo session on generic visuals — abstract motion, environment shots, hands at work, screen close-ups. Tag them. The library cuts an hour out of every future piece.
Voice clone for scale and continuity.
ElevenLabs voice clone of your founder/host removes the "is this AI?" question. Use it for narration only — keep one human take per piece for the hero line. Mix the two and you sound human without paying the studio bill.
A/B the hook, never the body.
The first 1.5 seconds is the only thing worth testing at this volume. Two hooks per piece, same body. Track watch-through to second 3. The winner becomes the template's hook template for next campaign.
The pipeline degrades by drift. Five symptoms that say the template is slipping — and the fix.
№ 01Veo shots all look the same.+
№ 02Runway adds artifacts.+
№ 03Audio feels off.+
№ 04Pipeline takes four days.+
№ 05Quality drops on export.+
Three drop-ins. The 15-second template, the Veo prompt scaffold, the ElevenLabs voice config.
The 15-second template.
Three beats, beat-locked cuts. The shape every piece fills.
15-SECOND TEMPLATE — beat breakdown ──────────────────────────────────── 00.0 → 01.5s HOOK - One visual that earns the next second. - Text on screen: 4-6 words max. - No setup. No "Hi, today we'll talk about..." 01.5 → 11.5s POINT - Three sub-beats, ~3.3s each. - Cut on the beat. Click track at 90 BPM. - Each sub-beat lands ONE specific claim, number, or name. 11.5 → 15.0s PAYOFF - The thing the viewer should feel/do/decide. - Outro card: brand mark, ONE CTA, no URL. GLOBAL - Caption font + size: locked across campaign. - Color grade: locked LUT. - Voice: single ElevenLabs voice, stability 0.5. - Aspect: 1080x1920, 30fps.
The Veo prompt scaffold.
Fill in the six axes. Generate three. Pick one. Move on.
VEO PROMPT — 6 axes, in this order SUBJECT: [one noun, concrete, specific] ACTION: [one verb, present tense] ENVIRONMENT:[one place, one time of day] CAMERA: [lens + move — e.g. 35mm, slow push-in] LIGHT: [source + quality — e.g. window backlight, soft] STYLE: [reference — e.g. cinematic verité, 35mm grain] NEGATIVE: [what to exclude — e.g. text, logos, faces] EXAMPLE (don't paste this — write your own): SUBJECT: a single brass key ACTION: spinning in mid-air ENVIRONMENT:dark wooden desk, late afternoon CAMERA: 50mm macro, slow rotation around the key LIGHT: single warm desk lamp, harsh shadow STYLE: cinematic, slight 35mm grain NEGATIVE: no hands, no text, no logos
The ElevenLabs voice config.
Pin the model. Lock the dial. Save it as a project default.
ELEVENLABS — campaign config
model_id: eleven_v3
voice_id: [your chosen voice — clone or library]
stability: 0.50 # do not drift
similarity: 0.80
style: 0.30 # subtle
speaker_boost: true
OUTPUT
format: mp3_44100_192
loudness: -16 LUFS (social)
-23 LUFS (broadcast / paid)
DON'T
- Change voice mid-campaign.
- Adjust stability per clip.
- Use the "creative" preset for narration.
Need this done for you? The author works on this exact thing with audit clients at austinaiguy.com.