How to Make an AI Short Film for Under $200: Full Production Workflow

What $200 and Two Days Can Actually Produce

Making a short film used to mean renting equipment, hiring crew, and booking locations. Now it means a laptop, a few AI subscriptions, and a willingness to generate a lot of clips you’ll never use.

This guide walks through the complete production workflow for an AI short film using Claude for script and prompt development, Seedance 2.0 for video generation, and Luma Canvas for assembly and finishing. It covers exact costs, honest shooting ratios, and the parts that will frustrate you.

The results won’t look like a Hollywood production. They’ll look like something else entirely — which is either a limitation or an aesthetic, depending on how you approach it.

Choosing Your Tools and Why These Three Work Together

There are dozens of AI video tools available right now. Picking three that work well together matters more than chasing the highest benchmark score.

Claude for Story and Prompt Development

Claude handles the language layer of this workflow. That means writing the script, yes, but more importantly it means turning story ideas into specific, structured prompts that video models can actually use.

Video generation models respond poorly to vague instructions. “A woman walks through a city” produces mediocre results. “A woman in her late 30s wearing a red wool coat walks slowly down a rain-wet cobblestone street at dusk, shallow depth of field, wide shot, slightly desaturated color grade” produces something you can cut into a film.

Claude’s Projects mode lets you maintain persistent context across sessions — your script, shot list, visual style guide, and character descriptions all stay loaded. This keeps your prompts consistent throughout production without having to re-paste reference material every time you open a new session.

Seedance 2.0 for Video Generation

ByteDance’s Seedance 2.0 produces high-quality clips at a price point that makes sub-$200 budgets viable. Key capabilities relevant to short film production:

Generates up to 10-second clips at 720p and 1080p
Handles motion physics well — objects fall naturally, water moves convincingly
Responds to camera motion prompts (pan, zoom, tracking shots) with reasonable accuracy
Faster generation times than some alternatives, which matters when you’re running 80+ clip attempts

The main limitation: character consistency across shots remains the hardest problem in AI video. Seedance 2.0 has improved on earlier models, but you’ll still manage this throughout production.

Luma Canvas for Assembly

Luma’s Dream Machine Canvas is a browser-based workspace for assembling, trimming, and refining AI video clips. It’s not a full NLE like Premiere or DaVinci Resolve, but it handles the specific needs of AI video production well — timeline editing, clip trimming, the ability to regenerate specific shots without leaving the workspace, basic color tools, and audio track import.

For more complex finishing, export your assembly cut and complete it in DaVinci Resolve, which is free and handles color grading professionally.

The Complete Budget Breakdown

Here’s the actual cost structure for a 2–3 minute short film on a two-day production window.

Tool	Plan/Tier	Approximate Cost
Claude.ai Pro	Monthly subscription	$20
Seedance 2.0	Credit pack (80–100 clip attempts)	$80–$110
Luma Dream Machine	Standard plan	$30
ElevenLabs	Starter plan (voiceover)	$5
Suno or Udio	Basic plan (music)	$8–$10
Upscaling (optional)	Online tool or Topaz	$0–$20
Total		$143–$195

Seedance 2.0 pricing is per second of video generated. A 10-second clip at 1080p costs roughly $0.80–$1.20 depending on resolution and motion complexity. If your shooting ratio is 4:1 and your finished film needs 25 clips, you’re generating around 100 attempts — plan credits accordingly.

If you already have an active Claude or Luma subscription, your out-of-pocket cost drops noticeably.

Pre-Production: Building the Story and Prompts with Claude

Don’t skip this stage. The quality of your prompts is the biggest variable in the entire workflow, and good prompts come from clear story logic.

Write a One-Page Treatment First

Before touching any video tool, spend an hour with Claude writing a one-page treatment. Include:

The central conflict or question — even a 2-minute film needs one
The setting and time period
The main character(s) and what they want
The emotional arc — what feeling should a viewer leave with

Use a prompt like this:

“I want to make a 2–3 minute AI short film. Here’s my initial idea: [idea]. Help me develop a one-page treatment with a clear story arc, a defined visual world, and a protagonist with a specific want. Keep it achievable with AI-generated footage.”

The “achievable with AI-generated footage” instruction matters. Claude will flag ideas that require consistent close-up character work across many locations (still difficult) or dialogue-heavy scenes (very difficult). Lean toward stories that work through visual metaphor, movement, and atmosphere.

Build a Shot List with Full Visual Descriptions

Once you have your treatment, ask Claude to generate a shot list. Each shot should include:

Shot type (wide, medium, close-up, aerial)
Subject and action
Setting and lighting conditions
Camera movement
Estimated duration
Emotional purpose of the shot

Ask Claude to format this as a table. You’ll use it as your prompt-generation reference throughout production.

Create a Visual Style Bible

Your style bible is a single document defining the visual language of your film. Every prompt you write will reference it.

Include:

Color palette and grade (e.g., “slightly desaturated, warm shadows, cool highlights, cinematic 2.39:1 aspect ratio”)
Lens aesthetic (e.g., “shallow depth of field, slight lens flare, 35mm film texture”)
Consistent time of day and weather conditions
Character descriptions — be specific about age, build, clothing, hair, distinguishing features

Ask Claude to write this document based on your treatment, then paste it into every subsequent session as context. This single step does more for clip consistency than any other part of the workflow.

Converting Shot Descriptions Into Video Prompts

This is where Claude earns its role in the pipeline. Take each shot from your list and ask Claude to write a Seedance-optimized prompt for it.

A prompt that works for Seedance 2.0 follows this structure:

[Subject] + [Action] + [Setting] + [Lighting] + [Camera movement and framing] + [Style/aesthetic keywords]

Example:

“An elderly man in a dark wool coat sits on a park bench feeding pigeons slowly, grey overcast morning light, medium shot, static camera, shallow depth of field, slightly desaturated film grain, melancholy atmosphere”

Claude can generate 25–30 of these in a single session when given your style bible and shot list as context. Building a full prompt library before you generate a single clip saves hours and reduces wasted credits.

Production: Generating Clips with Seedance 2.0

With prompts ready, you move into generation. This is where the budget gets spent.

Understanding the Shooting Ratio

This is the most important number to internalize before you start: expect a shooting ratio of 3:1 to 5:1.

For every usable clip in your finished film, you’ll generate 3–5 attempts. Sometimes more for shots involving specific character action or complex motion.

Common reasons clips don’t make the cut:

Unnatural limb movement or visual artifacts
Character appearance drifts — different face structure, different clothing details
Motion blur that obscures the subject
Background that conflicts with your style bible
The action happens too fast or too slow for your intended edit

Budget your generation credits with this ratio as your baseline. A 3-minute film needs roughly 25–35 distinct clips. At a 4:1 ratio, that’s 100–140 total generations.

Prompt Patterns That Improve Output Quality

A few techniques that consistently produce better results in Seedance 2.0:

Anchor in the middle of the action. Instead of prompting a character to begin walking, prompt them already mid-stride. AI video models continue action more naturally than they initiate it from rest.

Specify what you don’t want. Adding exclusions helps: “no text overlay, no watermark, no lens distortion, no sudden camera shake.”

Use cinematographic vocabulary. Terms like “tracking shot,” “dolly zoom,” “handheld,” “rack focus,” and “hero shot” are in the training data and reliably produce the described effect.

Limit motion complexity in close-ups. Close-up shots with significant movement are harder to generate cleanly. Complex motion works better in wide and medium shots.

Managing Character Consistency

Character consistency is the main technical challenge in this workflow. Practical approaches:

Reduce character screen time. Design your story around environment shots, object close-ups, silhouettes, and atmospheric shots rather than full-face coverage. This is a legitimate directorial choice.

Use reference images. Seedance 2.0 accepts image references. Take a generation you like and feed it as a reference for subsequent shots featuring the same character. Consistency improves substantially.

Accept variation as an aesthetic. Some AI films lean into the slightly shifting, dreamlike quality of inconsistent character rendering. If your story has surreal or psychological elements, this can work for you rather than against you.

Dedicate an early session to character lock-in. Spend 10–15 generations finding your character’s look before generating any story clips. Lock in the best result and use it as your reference throughout.

Organizing Your Output

Rename clips by scene and shot number matching your shot list (scene_02_shot_04_take_3.mp4) as you download them. The temptation to sort this out later is real and regrettable.

Keep all generations, including failed ones. A clip that doesn’t work for its intended shot sometimes cuts perfectly into a different part of the film.

Post-Production: Assembling in Luma Canvas

With 80–140 clips generated, you’ll have 25–35 usable ones. Now you build the film.

The Assembly Cut

Import your usable clips into Luma Canvas and do a rough assembly cut first — put clips in story order without worrying about timing, transitions, or audio. Watch it through once.

This first watch immediately shows you:

Where the visual logic breaks between clips
Pacing problems — sequences that drag, moments that need more coverage
Coverage gaps — shots you forgot to generate

Make a list of pickups. Budget one additional generation session (10–15 clips) for these. Every film needs them.

Editing Principles for AI Footage

AI video has specific rhythms that differ from live-action. A few principles:

Cut on motion. Transitions feel smoother when both outgoing and incoming clips have visual movement. Static-to-static cuts often feel abrupt with AI footage.

Use long clips at near-full length. If Seedance gives you a great 8-second clip, use 7 of those seconds. AI footage benefits from extended holds in a way that live-action sometimes doesn’t.

Color before continuity. AI clips from the same prompt set will have slight color variations. Apply a global base grade in Luma Canvas or DaVinci Resolve before you finalize cut continuity.

Build rhythm against sound. Import a scratch music track before finalizing your cuts. Sound shapes how long each shot should hold more than any visual logic will.

Audio Is Where the Film Lives

Audio does more heavy lifting in AI short films than in live-action. It’s also where you spend the least and gain the most.

Narration or dialogue: ElevenLabs produces natural-sounding voiceover at the Starter tier. Write your narration script in Claude, refine the voice in ElevenLabs, and layer it under your visuals.

Music: Suno and Udio generate original music to a specified style and duration. Describe emotional tone, instrumentation, and length: “melancholy ambient piano piece, 2 minutes 30 seconds, sparse, appropriate for a short film montage.” Generate 5–6 options and pick the best two.

Sound design: Freesound.org has a large, free library of licensed sound effects. Even minimal ambient layering — city noise, wind, footsteps — transforms the viewing experience significantly.

Final Export

Luma Canvas exports up to 1080p. For 4K delivery, run your export through a video upscaler. Master your audio before final export — peaks at -1 dBFS, dialogue sitting around -12 dBFS — and your film will hold up on any playback system.

Automating This Workflow with MindStudio

The workflow above involves a lot of manual context-switching: Claude to Seedance to Luma, managing prompt files, tracking clip versions, running pickups. Once you’ve been through it once, you start looking for ways to streamline it.

MindStudio’s AI Media Workbench brings the generation and post-production layer of this workflow into a single workspace. Instead of maintaining separate accounts and switching between tabs, you access Seedance 2.0, image generation models for reference images, upscaling tools, and clip merging utilities in one place — no API keys or separate account setup required.

More usefully, MindStudio’s no-code workflow builder lets you chain these tools into automated pipelines. A basic automation for this workflow might look like:

Accept a shot description as input
Pass it through Claude to generate an optimized prompt against your style bible
Send that prompt to Seedance 2.0 for generation
Apply an upscaling step automatically
Output the finished clip to a designated folder

That’s a standard multi-step workflow in MindStudio — buildable in under an hour. For anyone making multiple films or working on series content, automating the prompt-to-clip pipeline meaningfully reduces the manual work per shot. You can also explore building AI-assisted content workflows that extend beyond video — handling script drafts, social distribution copy, thumbnail generation, and more in a single agent.

MindStudio also provides access to 200+ AI models, so when a better video generation model releases — and one will — you update your workflow without rebuilding the pipeline from scratch.

You can try MindStudio free at mindstudio.ai.

Common Mistakes That Waste Budget and Time

Over-scoping the story. A 5-minute film with 4 characters across 8 locations is not a 2-day, $200 project. A 2-minute film following one character through one environment is. Constraint is your asset, not your limitation.

Skipping the style bible. If you don’t define your visual language before generating, you’ll spend your budget creating inconsistent clips that don’t cut together. The style bible takes 30 minutes to write and saves hours of regeneration.

Not testing one clip per scene first. Before running all takes for a scene, generate a single test clip. If it doesn’t match your style bible, refine the prompt before spending credits on 4 more attempts.

Expecting every generation to succeed. A 3:1 shooting ratio means 2 out of 3 clips don’t make the film. That’s not failure — that’s how the format works. Budget for it upfront so it doesn’t feel like a problem when it happens.

Underinvesting in audio. Mediocre visuals with good audio feel professional. Good visuals with mediocre audio feel cheap. The audio tools in this workflow are among the least expensive parts of the budget and among the highest-return.

Frequently Asked Questions

How long does it actually take to make an AI short film?

Two focused days is realistic for a 2–3 minute film. Day one covers pre-production — treatment, shot list, style bible, and prompt generation with Claude — plus the first round of clip generation. Day two covers editing, audio, pickups, and finishing in Luma Canvas. More complex films with higher coverage needs extend to 4–5 days.

What is a shooting ratio and why does it matter for AI video production?

A shooting ratio is the total number of clips generated divided by the number you actually use in the final film. In live-action, 10:1 is common. In AI video generation, expect 3:1 to 5:1 on average. Understanding your ratio before you start lets you budget generation credits accurately and removes the frustration of clips that don’t work — it’s part of the process, not a failure.

How do you maintain character consistency across AI-generated video clips?

Use reference images — take a generation you like and feed it as the visual anchor for subsequent shots with the same character. Design your story to minimize full-face character coverage, relying more on environment shots, silhouettes, and reaction shots. You can also spend a dedicated early generation session finding and locking your character’s visual appearance before generating any narrative clips.

How much does it cost to make a 3-minute AI short film?

Using the workflow in this article, budget $140–$200. The largest single cost is video generation credits, and your shooting ratio directly affects this — a 5:1 ratio costs roughly twice as much in generation credits as a 3:1 ratio. Tighter, more specific prompts tend to produce better first-attempt results, which lowers your average cost per usable clip over time.

Can AI-generated short films be distributed or monetized?

Most AI video platforms’ terms permit commercial use of generated outputs for paid subscribers — read the specific terms for each tool you use. Music generated by tools like Suno has more nuanced licensing; review current terms before publishing to platforms with Content ID systems like YouTube or TikTok. For festival submission, many film festivals now accept AI-assisted work, particularly in experimental and short film categories.

Is Claude the best tool for writing AI video prompts?

Claude performs well here because it takes detailed style instructions seriously and maintains context consistently across long sessions in Projects mode. GPT-4o and Gemini produce comparable results for individual prompt generation. Claude’s advantage in this specific workflow is the ability to hold your entire style bible, shot list, and character descriptions as persistent context — which keeps your prompts visually consistent across a full production session rather than drifting between conversations.

Key Takeaways

A complete AI short film workflow using Claude, Seedance 2.0, and Luma Canvas is achievable for under $200 when planned carefully.
Pre-production — treatment, shot list, style bible, and optimized prompts — is the highest-leverage stage. Do it before spending a dollar on generation.
Budget for a 3:1 to 5:1 shooting ratio. It’s not waste; it’s how AI video production works at current model capability.
Audio does disproportionate work in AI films. Invest time in voiceover, music, and ambient sound even when the rest of the budget is tight.
Tools like MindStudio’s AI Media Workbench can automate the cross-tool parts of this workflow — worthwhile if you’re making multiple films or need to reduce per-clip production time.

The barrier to making a competent AI short film is now primarily creative and organizational, not financial. The workflow is learnable in a single production cycle. What you make with it is the part that’s still entirely up to you.