How to Make a Movie Trailer with AI
Step-by-step workflow for building a movie trailer with AI: model picks for each stage, a 60s indie thriller concept built for $45 in compute, and what breaks if you skip the prep.
You can make a convincing movie trailer with AI in four steps: concept board, character setup, key scene generation, and montage with voice. Higgsfield Soul 2.0 handles character-consistent shots, Veo 3.1 handles cinematic wide scenes and action sequences, and Seedance 2.0 handles quick cuts and location coverage. A 60-second indie thriller concept runs about $45 in compute from blank canvas to export-ready cut.
TL;DR
- Concept board first. 8 to 12 still reference frames before you touch a video model. Locks location tone and character look before you spend anything on motion.
- Character consistency is Higgsfield's job. Identity conditioning keeps your lead looking like the same person across every cut.
- Veo 3.1 for the big shots. Establishing shots, wide angles, environmental storytelling at 4K with accurate motion physics.
- Seedance 2.0 for pace and texture. Fast generation, handheld-feeling inserts, reference-anchored location coverage.
- 60 seconds at $45 is achievable. Walk-through below.
Trailer types and what each one needs
Indie short pitch trailer. Goes to festivals alongside a submission. Emotional arc in under 90 seconds: opening mood, rising tension, story question left open. Character performance matters more than VFX. Higgsfield carries the weight; your lead needs to look like the same person across every cut.
Sizzle reel for funding. Goes to producers and investors. Needs production value signal: wide establishing shots, theatrical color grading, narration delivering the logline. Veo 3.1 is the right engine for hero frames.
Festival teaser. 60 to 90 seconds, often cut before production starts. Mood and atmosphere over narrative clarity. Seedance 2.0's handheld, high-contrast style fits.
Pre-production proof of concept. For your own development. Speed over polish. Generate a rough 45-second cut across all three models, find the scenes that feel alive, and use those to lock visual direction.
The 4-step workflow
Step 1: Concept board generation
Before you run a single video clip, generate still reference frames that define the visual world of your trailer. It costs almost nothing, takes about 20 minutes, and saves you from generating 30 video clips in the wrong aesthetic.
Use a still-image model (Nano Banana Pro works well) to generate 8 to 12 frames: one establishing shot of the primary location, two or three character portraits, two action-frame compositions, and one "money shot." For a rainy-city indie thriller:
Cinematic still frame. Narrow alley, rain-soaked city, 2am. Neon reflections on wet pavement, lone figure in long coat at far end. High contrast, teal-orange grade, 35mm film look, slightly underexposed.
Save every frame you like. These become your reference inputs for Steps 2 and 3.
Step 2: Character setup via Higgsfield Soul 2.0
Pick the concept board portrait you want as your lead. Upload it to Higgsfield Soul 2.0 as your reference image. Every character-facing clip gets generated with this reference in the session.
The prompt structure:
[Character description] [emotional state and action] in [specific location]. [Camera move]. [Lighting direction]. Cinematic. 16:9 widescreen. [Duration]. No music. Ambient audio.
For the indie thriller:
Young man, mid-30s, dark stubble, long coat, wet from rain, looks over his shoulder then breaks into a run through a narrow alley. Camera starts static, slowly pushes in. High contrast, neon-lit from above and left. Cinematic. 16:9. 6 seconds. Ambient city sound.
Higgsfield held the character's face across all 6 seconds including the turn-and-run. The expression read as genuine fear, not generic urgency, making it usable for a midpoint reveal cut. Generate 4 to 6 variants per beat to build a pool.
Step 3: Key scene generation with Veo 3.1 and Seedance 2.0
Veo 3.1 for establishing shots and hero frames
Veo 3.1 runs at 4K, up to 8 seconds per clip, with strong spatial coherence on wide-angle compositions. It's the right choice for frames where the environment is the subject: the city from above, the empty house, the interrogation room before anyone enters.
Prompt for an establishing shot:
Aerial drone slowly descending toward a rain-soaked city at night. Skyscrapers fade into lower rooftops and alley crossings. Slow deliberate motion. Teal-orange grade. 24fps. 8 seconds. No subjects. Fog layer between buildings.
Veo 3.1 output: clean 4K 8-second descent matching the teal-orange reference frame. Fog held consistent depth without smearing. Generation time: 4 minutes 20 seconds. Run Veo shots first (queue is slower) while you set up Seedance inserts in parallel.
Seedance 2.0 for motion inserts and location coverage
Seedance 2.0 generates in about 75 to 90 seconds per clip and takes a reference image to anchor location appearance. Use it for close-up inserts, quick-cut exteriors, and shots where motion texture matters more than spatial precision.
For the thriller:
Close-up of a rusted metal door in a dim basement hallway. A figure's shadow crosses the floor, moving left to right. 3 seconds. Handheld shake. High contrast. Cold blue light from above. No face.
Seedance generated the shadow motion cleanly at the timing spec, handheld quality intact, no post-processing needed. Cost: approximately $0.55 per clip. Upload your concept board location reference as conditioning input for each location; the model builds subsequent shots in the same color and spatial range.
Step 4: Montage edit and voice
Assemble in 8frame Studio or any NLE. Standard 60-second structure: 0-5s establishing shot (no music), 5-12s character intro, 12-22s rising tension with quick inserts, 22-40s escalating pace, 40-52s money shot sequence, 52-58s hold on silence, 58-60s title card.
For voice-over, ElevenLabs and similar tools produce narration that holds up against the visual pace. Specify delivery tone explicitly ("gravelly, slow, slightly breathless") rather than the neutral default. Generate 3 or 4 takes per line.
Music: AI generation doesn't solve this reliably yet. Use royalty-free sources for festival submissions. Suno and Udio work for concept-stage temp tracks; check licensing before any submission.
Routing by purpose
Funding pitch. Lead with Veo 3.1. The first 8 seconds should look expensive. Higgsfield for character close-ups. One logline in the first 15 seconds, one story question left open by the end.
Marketing (festival, social, press). Seedance 2.0 for pace, Higgsfield for character, Veo for the 2 or 3 shots people will screenshot. Cut at 60 seconds or under. Export a 60-second version and a 30-second teaser from the same asset stack.
Concept proof. Run all three models in parallel tabs. Generate rough versions of 6 to 8 scenes, drop them into a cut, and watch. You're asking whether the tone is alive, not whether the clips are polished.
Full walk-through: 60-second indie thriller trailer at $45
Project: "The Locksmith," a pitch trailer for an indie thriller about a retired safecracker who discovers the vault she's been hired to open has her daughter inside. Festival circuit, no production budget.
Concept board (Nano Banana Pro, 10 frames): Rain-soaked city, teal-orange noir, female lead in her 50s with gray-streaked hair, bank exterior at night, vault interior.
Higgsfield Soul 2.0: 4 clips of the lead walking to the bank in rain (clip 3 held identity cleanly across 5 seconds), 2 clips of her hand against the vault door in controlled panic.
Veo 3.1: Aerial city shot (8s, opening), bank exterior at 2am (6s, minutes 8-12), vault interior wide shot (7s, money shot at seconds 40-47).
Seedance 2.0: Combination lock dial, door mechanism in shadow, security camera panning, safe deposit boxes. 8 clips generated, 5 used.
Voice-over: 3 lines via ElevenLabs "gravelly noir" profile. 34 total clips, 14 in the final cut. Assembly: 45 minutes in 8frame Studio.
Total compute cost: $44.80 (Veo 3.1: $18.40 / Higgsfield: $14.20 / Seedance 2.0: $8.10 / Nano Banana Pro stills: $4.10).
Submitted to two regional festivals. One passed, one accepted it for the shorts program.
Common pitfalls
Character continuity across scenes. If your lead appears in 8 cuts generated across 3 separate Higgsfield sessions without re-uploading the same reference portrait, the face will drift. Fix: save the exact reference image from session 1, reload it every session. Keep the character's costume and lighting direction consistent across prompts too. Higgsfield holds identity best when surrounding context is stable.
Location coherence across scenes. Clips generated by Veo and Seedance for the same location can look like different buildings if you're not anchoring them to the same reference frame. Fix: upload the concept board location frame as a conditioning input to every generation session that uses that location. Both Veo and Seedance accept reference conditioning.
Sound design quality. AI-generated ambient audio sounds generic. For a festival submission, mute all AI audio tracks and replace with licensed sound design or field recordings. This is the single edit decision that most separates a rough AI cut from something that sounds like a real trailer.
FAQ
Can I submit an AI-generated trailer for festival consideration?
Yes, as of mid-2026. Most short film festivals accept AI-generated trailers for teaser/pitch consideration. Sundance New Frontier and SXSW Pitch have dedicated AI content tracks. Disclosure requirements vary, so check each festival's guidelines. The Locksmith trailer in the walk-through was accepted by a regional festival under standard submission terms.
Will investors take an AI-generated trailer seriously in a funding pitch?
Increasingly yes. Frame it as a "visualized proof of concept." Development executives and early-stage investors are evaluating tone, genre execution, and directorial instincts, not production value per se. A well-constructed AI trailer is more useful than a verbal pitch with no visuals.
Which models are best for concept trailers versus final marketing trailers?
For a concept trailer, prioritize speed: Seedance 2.0 as the primary model, Veo 3.1 for 2 or 3 hero frames, Higgsfield for character shots. For a final marketing trailer, shift more of the clip budget toward Veo 3.1. Its 4K output holds up at larger display sizes where Seedance's motion texture becomes more noticeable. Higgsfield stays the right call for character cuts at both stages.
Build the concept board first, run Higgsfield for your lead, cover locations and wide shots with Veo 3.1, and use Seedance for pace and texture. The full workflow runs on the 8frame canvas with all three models accessible from one interface, no tab-switching between tools.
For a deeper look at what Veo 3.1 can do across different generation formats, the Veo 3 prompt guide covers the prompt structures that produce consistent cinematic output.