Workflow recipes by category
TL;DR: Pick the recipe that matches your job. Each one gives you a starting model type, the fastest first pass, the main quality lever, the main cost lever, and when to finish in Premiere instead of generating more.
How to use these recipes
Section titled “How to use these recipes”These are starting points, not rules. For each recipe, open modelBridge’s Browse tab, filter by the suggested category, and check the cost badge and time estimate on the Generate button before committing to a model. Click ⓘ on any field for guidance on what it does.
Social 9:16 loop
Section titled “Social 9:16 loop”Goal: Short, eye-catching loop for TikTok, Instagram Reels, YouTube Shorts.
In modelBridge: Filter Browse by Text-to-video or Image-to-video. Look for models with lower cost badges and shorter time estimates.
Fastest first pass: 2–3 seconds, vertical aspect ratio, single subject, simple motion.
Main quality lever: Prompt simplicity. One focal point, one motion direction.
Main cost lever: Duration. Every extra second costs proportionally more. The cost badge updates in real time as you adjust duration — check it before generating.
Common trap: Overloading the frame. Text overlays, multiple subjects, complex backgrounds all fail at 9:16.
Self-learning at work: If a previous generation failed due to wrong aspect ratio or file size, modelBridge caught and saved that constraint. Your next attempt on the same model won’t make the same mistake.
Finish in Premiere:
- Beat-match → Window → Essential Sound, use Remix to match audio to clip length
- Text and captions → Window → Essential Graphics, use vertical text safe zones
- Color pop → Window → Lumetri Color, push saturation and contrast for feed visibility
Product explainer shot
Section titled “Product explainer shot”Goal: Show a product’s function, USP, or detail clearly.
In modelBridge: Filter Browse by Image-to-video. Select your product photo or mockup as source. Use the preview panel to evaluate the result before it touches your timeline.
Fastest first pass: Static product with subtle motion — gentle rotation, soft zoom, ambient movement.
Main quality lever: Source image quality. A clean, well-lit product photo gives the model less to invent.
Main cost lever: Generate the hero moment only — 3–4 seconds. Cut together multiple short clips in Premiere rather than one long expensive generation.
When Import to Timeline: modelBridge replaces your source product photo on the timeline at the exact same position and scale. Without modelBridge: delete source, import generated file, drag to exact timecode, scale manually.
Common trap: Asking the model to show product functionality. Show the product, explain function with real graphics in Premiere.
Finish in Premiere:
- UI overlays → Window → Essential Graphics, place on V2 above the generated clip
- Product labels → generate background plate only in modelBridge, add real text in Essential Graphics
- Brand fonts → File → New → Legacy Title or Essential Graphics
Mood film and concept teaser
Section titled “Mood film and concept teaser”Goal: Establish atmosphere, emotion, or visual language for a campaign or pitch.
In modelBridge: Filter Browse by Text-to-video. Use Dual Mode to compare two model families before committing — aesthetic differences between families are most visible in mood work.
Fastest first pass: Abstract or environmental shots first — light, texture, color, motion. No people, no text.
Main quality lever: Style descriptors in your prompt. “Film grain, anamorphic lens flare, shallow depth of field, muted palette” shapes output more than subject description.
Main cost lever: Generate at shorter duration first. Check the cost badge before each generation.
Common trap: Over-specifying too early. Vary seeds across a few generations before locking direction.
Finish in Premiere:
- Transitions → Effects panel → Video Transitions → Dissolve → Film Dissolve
- Color grade → Window → Lumetri Color, pull shadows toward a tint
- Cut to music first, then generate clips to fit the rhythm
Title and typography sequences
Section titled “Title and typography sequences”Goal: Opening titles, chapter cards, lower thirds, branded sequence backgrounds.
In modelBridge: Filter Browse by Text-to-video or Image-to-video. Abstract and texture-focused models work best. Avoid any model that tries to render text.
Fastest first pass: Generate the background element only — moving texture, abstract shapes, light leak. No text in the prompt, ever.
Main quality lever: Motion complexity. Slow, fluid motion holds better behind titles than fast chaotic movement.
Main cost lever: Reuse one generated background across multiple title cards. Generate once, cut into sections in Premiere.
Always: Add all real text in Premiere. Never ask the model to generate readable type.
Finish in Premiere:
- Title text → Window → Essential Graphics, use brand font, set safe margins
- Lower thirds → build as Motion Graphics Template for reuse
- Kinetic timing → Effect Controls → Motion → Position keyframes for text entry and exit
Rescue a bad archive clip
Section titled “Rescue a bad archive clip”Goal: Make a weak or damaged shot usable — noisy, low resolution, wrong framing, or poorly lit.
In modelBridge: Filter Browse by Upscaling or Video-to-video. Start with upscaling — lowest risk. Preview the result in modelBridge before importing.
Fastest first pass: Upscaling first, always.
Main quality lever: The original clip itself. AI amplifies what’s there — it can’t invent missing detail.
Main cost lever: Upscaling is cheaper than video-to-video enhancement. Fix with upscaling first.
Common trap: Over-processing. One intervention only. The self-learning validation system remembers format and duration constraints — your second attempt never fails on the same technical issue.
When to stop using AI: Wrong content — wrong framing, wrong action — AI can’t fix that. Cut around it or find a replacement.
Fix in Premiere:
- Noise reduction → Effects panel → Video Effects → Noise & Grain → Median
- Brightness → Window → Lumetri Color → Basic Correction
- Reframe → Effects panel → Video Effects → Transform → Crop, then Effect Controls → Motion → Scale
Voice-led sequence
Section titled “Voice-led sequence”Goal: Generate visuals that support a voiceover, or replace a scratch VO with a polished AI voice.
In modelBridge: Filter Browse by Text-to-video or Image-to-video for visuals. For voiceover generation or replacement, filter by Audio — ElevenLabs models are here.
Fastest first pass: Lock your audio first. Record or import the voiceover track in Premiere before generating a single visual. Generate visuals to fit the edit — not the other way around.
Main quality lever: Script specificity. Brief each prompt to the exact sentence it illustrates.
Main cost lever: Reuse establishing shots across multiple lines. One strong environment shot can cover 3–4 seconds of narration.
For TTS or voice replacement: Select an audio model, generate, preview in modelBridge before touching the timeline. Generated audio routes automatically to the first available audio track on import. Tag to client/project before generating.
Common trap: Generating visuals before locking audio. You’ll chase the voiceover for the rest of the edit.
Finish in Premiere:
- Sync visuals to voiceover → Window → Essential Sound, tag as Dialogue to auto-duck music under voice
- B-roll pacing → Razor tool (C), cut visuals to breath points
- Captions → Window → Text → Transcript (Premiere 2023+)
Your personal workflow menu
Section titled “Your personal workflow menu”Pick one or two recipes per project. Don’t over-complicate it.
The fastest editors aren’t the ones who know every model — they’re the ones who know which recipe to start with, and when to stop iterating and finish in Premiere.
Common failure modes — when something in your recipe isn’t working.
From moodboard to locked shot — how to run a full generation pass from brief to locked clip.
Building a signature look — how to build a consistent visual system across a project.